Context Switching with Offload Processors

A method for context switching of multiple offload processors is disclosed. The method can include receiving network packets for processing through a memory bus connected socket, organizing the network packets into multiple sessions for processing, suspending processing of at least one session by reading a cache state of at least one of the offload processor into a context memory by operation of a scheduling circuit, with virtual memory locations and physical cache locations being aligned, and subsequently directing transfer of the cache state to at least one of the offload processors for processing by operation of the scheduling circuit.

Skip to: Description  ·  Claims  · Patent History  ·  Patent History

Description

PRIORITY CLAIMS

This application claims the benefit of U.S. Provisional Patent Applications 61/753,892 filed on Jan. 17, 2013, 61/753,895 filed on Jan. 17, 2013, 61/753,899 filed on Jan. 17, 2013, 61/753,901 filed on Jan. 17, 2013, 61/753,903 filed on Jan. 17, 2013, 61/753,904 filed on Jan. 17, 2013, 61/753,906 filed on Jan. 17, 2013, 61/753,907 filed on Jan. 17, 2013, and 61/753,910 filed on Jan. 17, 2013, the contents all of which are incorporated by reference herein.

TECHNICAL FIELD

Described embodiments relate to deterministic context switching for computer systems that include a memory bus connected module with offload processors.

BACKGROUND

Context switching (sometimes referred to as a process switch or a task switch) is the switching of a processor from execution of one process or thread to another. During a context switch the state (context) of a process is stored in memory so that execution can be resumed from the same point at a later time. This enables multiple processes to share a single processor and support a multitasking operating system. Commonly, a process is an executing or running instance of a program that can run in parallel and share an address space (i.e., a range of memory locations) and other resources with their parent processes. A context generally includes the contents of a processor's registers and program counter at a specified time. An operating system can suspend the execution of a first process and store the context for that process in memory, while subsequently retrieving the context of a second process from memory and restoring it in the processor's registers. After terminating or suspending the second process, the context of the first process can be reloaded, resuming execution of the first process.

However, context switching is computationally intensive. A context switch can require considerable processor time, which can be on the order of nanoseconds for each of the tens or hundreds of context switches per second. Since modern processors can handle hundreds or thousands separate processes, the time devoted to context switching can represents a substantial cost to the system in terms of processor time. Improved context switching methods and systems can greatly improve overall system performance and reduce hardware and power requirements for server or other data processing systems.

SUMMARY

This disclosure describes embodiments of systems, hardware and methods suitable for context switching of processors in a system. Embodiments can include multiple offload processors, each connected to a memory bus, with the respective offload processors each having an associated cache with an associated cache state. A low latency memory can be connected to multiple offload processors through the memory bus, and a scheduling circuit can be used for directing storage of a cache state from at least one of the respective offload processors into the low latency memory, and for later directing transfer over the memory bus of the cache state to at least one of the respective offload processors. In certain embodiments, including those associated with the use of ARM architecture processors, the multiple offload processors can have an accelerator coherency port for accessing cache state with improved speed. In other embodiments, a common module can support the offload processors, low latency memory, and scheduling circuit, with access for external network packets provided through a memory socket mediated connection, including but not limited to dual in line memory module (DIMM) sockets.

In some embodiments the associated cache state includes at least one of: a state of the processor registers saved in register save area, instructions in the pipeline being executed, stack pointer and program counter, prefetched instructions and data waiting to be executed by the session, and data written into the cache. The system can further include an operating system (OS) running on at least one the multiple offload processors. The OS and scheduling circuit can cooperate to establish session contexts that are physically contiguous in a cache. Session color, size, and starting physical address can be communicated to the scheduler circuit upon session initialization and a memory allocator used to determine a starting address of each session, the number of sessions allowable in the cache, and the number of locations wherein a session can be found for a given color.

According to embodiments, a cache state stored by one of the multiple offload processors can be transferred to another offload processor. In particular applications, this can enable a scheduling circuit to prioritize processing of network packets in a first queue received over the memory bus by stopping a first session associated with one of the offload processors, storing the associated cache state, and initiating processing of network packets held in a second queue.

Embodiment can also include a method for context switching of multiple offload processors, each having an associated cache with an associated cache state, and using a low latency memory connected to multiple offload processors through the memory bus. The method includes directing storage of a cache state via a bulk read from at least one of the respective offload processors into the low latency memory using a scheduling circuit, with any virtual and physical memory locations being aligned. Subsequently, transfer is directed over the memory bus of the cache state to at least one of the respective offload processors for processing, with transfer being controlled by the scheduling circuit. As with the structure embodiments previously described, common module can support the offload processors, low latency memory, and scheduling circuit, with access for external network packets provided through a DIMM or other memory socket connection.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1-0 shows a system having context switching according to an embodiment.

FIG. 1-1 is a diagram showing page collisions in a physically indexed cache without coloring.

FIG. 1-2 shows a virtually indexed cache.

FIG. 1-3 shows a virtual/physical aligned cache according to an embodiment.

FIGS. 2-0 to 2-3 show processor modules according to various embodiments.

FIG. 2-4 shows a conventional dual-in-line memory module.

FIG. 2-5 shows a system according to another embodiment.

FIG. 3 shows a system with a memory bus connected offload processor module having context switching capabilities, according to one embodiment.

FIG. 4 is a flow diagram showing context switching operations according to one particular embodiment.

DETAILED DESCRIPTION

Various embodiments will now be described in detail with reference to a number of drawings. The embodiments show modules, systems and methods for switching contexts in offload processors that are connected to a system memory bus. Such offload processors can be in addition to any host processors connected to the system memory bus, and can operate on data transferred over the system memory bus independent of any host processors. In particular embodiments, offload processors can have access to a low latency memory, which can enable rapid storage and retrieval of context data for rapid context switching. In very particular embodiments, processing modules can populate physical slots for connecting in-line memory modules (e.g., dual in line memory modules (DIMMs)) to a system memory bus.

FIG. 1-0 shows a system 100 according to one embodiment. A system 100 can include one or more offload processors 118, a scheduler 116, and a context memory 120. An offload processor 118 can include one or more processor cores that operate in conjunction with a cache memory. In a context switch operation, the context of a first processing task of offload processor 118 can be stored in context memory 120, and the offload processor 118 can then undertake a new processing task. Subsequently, the stored context can be restored from the context memory 120 to the offload processor 118, and the offload processor 118 can resume the first processing task. In particular embodiments, the storing and restoring of context data can include the transfer of data between the cache of an offload processor 118 and the context memory 120.

A scheduler 116 can coordinate context switches of offload processors 118 based on received processing requests. Accordingly, a scheduler 116 can be informed of, or can have access to, the state of offload processors 118 as well as the location of context data for the offload processors 118. Context data locations can include locations in processor cache, as well as locations in context memory 120. A scheduler 116 can also follow, or be updated with, a state of offload processors 118.

As understood from above, a context memory 120 can store context data of offload processors 118 for subsequent retrieval. A context memory 120 can be separate from cache memories of the offload processors. In some embodiments, a context memory 120 can be low latency memory as compared to other memory in the system, to enable rapid context storage and retrieval. In some embodiments, a context memory 120 can store data other than context data.

In the particular embodiment shown, offload processors 118, scheduler 116, and context memory 120 can be part of a module 122 connected to a memory bus 124. Data and processing tasks for execution by offload processors 118 can be received over memory bus 124. In some embodiments, transfers of context data between offload processors 118 and context memory 120 can occur over memory bus 124. However, in other embodiments, such transfers can occur over a different data path on the module 122.

Referring still to FIG. 1-0, in the very particular embodiment shown, a method can further include a second switch 114, a memory controller 112, a host processor 110, an input/output (I/O) fabric 108, and a first switch 106. A second switch 114 can be included on module 122. The particular system 100 of FIG. 1-0 is directed to network packet processing scheduling and traffic management, but it is understood that other embodiments can include context switching operations, as described herein or equivalents, directed to other types of processing tasks.

In the particular embodiment of FIG. 1-0, a first switch 106 can receive and/or transmit data packets 104 from data source 102. A data source 102 can be any suitable source of packet data, including the Internet, a network cloud, inter- or intra-data center networks, cluster computers, rack systems, multiple or individual servers or personal computers, or the like. Data can be packet or switch based, although in particular embodiments non-packet data is generally converted or encapsulated into packets for ease of handling. The data packets typically have certain characteristics, including transport protocol number, source and destination port numbers, or source and destination (Internet Protocol) IP addresses. The data packets can further have associated metadata that helps in packet classification and management.

A switch 106 can be a virtual switch (an I/O device). A switch 106 can include, but is not limited to, devices compatible with peripheral component interconnect (PCI) and/or PCI express (PCIe) devices connecting with host motherboard via PCI or PCIe bus 107. The switch 106 can include a network interface controller (NIC), a host bus adapter, a converged network adapter, or a switched or an asynchronous transfer mode (ATM) network interface. In some embodiments, a switch 106 can employ IO virtualization schemes such as a single root I/O virtualization (SR-IOV) interface to make a single network I/O device appear as multiple devices. SR-IOV permits separate access to resources among various PCIe hardware functions by providing both physical control and virtual functions. In certain embodiments, the switch 106 can support OpenFlow or similar software defined networking to abstract out of the control plane. The control plane of the first virtual switch performs functions such as route determination, target node identification etc.

A switch 106 can be capable of examining network packets, and using its control plane to create appropriate output ports for network packets. Based on route calculation for the network packets or data flows associated with the network packets, the forwarding plane of the switch 106 can transfer the packets to an output interface. An output interface of the switch may be connected with an IO bus, and in certain embodiments the switch 106 may have the capability to directly (or indirectly, via an I/O fabric 108) transfer the network packets to a memory bus interconnect 109 for a memory read or write operation (direct memory access operation). Functionally, for certain applications the network packets can be assigned for transport to specific memory locations based on control plane functionality.

Switch 106, connected to an IO fabric 108 and memory bus interconnect 109, can also be connected to host processor(s) 110. Host processor(s) 110 can include one or more host processors which can provide computational services including a provisioning agent 111. The provisioning agent 111 can be part of an operating system or user code running on the host processor(s) 110. The provisioning agent 111 typically initializes and interacts with virtual function drivers provided by system 100. The virtual function driver can be responsible for providing the virtual address of the memory space where a direct memory addressing (DMA) is needed. Each device driver can be allocated virtual addresses that map to the physical addresses. A device model can be used to create an emulation of a physical device for the host processor 110 to recognize each of the multiple virtual functions (VF) that can be created. The device model can be replicated multiple times to give the impression to VF drivers (a driver that interacts with a virtual IO device) that they are interacting with a physical device. For example, a certain device model may be used to emulate a network adapter that the VF driver can act to connect. The device model and the VF driver can be run in either privileged or non-privileged mode. There can be no restriction with regard to which device hosts/runs the code corresponding to the device model and the VF driver. The code, however, can have the capability to create multiple copies of device model and VF driver so as to enable multiple copies of said I/O interface to be created. In certain embodiments the operating system can also create a defined physical address space for applications supported by VF drivers. Further, the host operating system can allocate a virtual memory address space to the application or provisioning agent. The provisioning agent 111 can broker with the host operating system to create a mapping between virtual addresses and a subset of the available physical address space. The provisioning agent 111 can be responsible for creating each VF driver and allocating it a defined virtual address space.

A second virtual switch 114 can also be connected to the memory controller 112 using memory bus 109. The second virtual switch 114 receives and switches traffic originating from the memory bus 109 both to and from offload processors 118. Traffic may include, but is not limited to, data flows to virtual devices created and assigned by the provisioning agent 111, with processing supported by offload processors 118. The forwarding plane of the second virtual switch transports packets from a memory bus 109 to offload processors 118 or from the offload processors 118 back onto the memory bus 109. For certain applications, the described system architecture allows relatively direct communication of network packets to the offload processors 118 with minimal or no interruptions to a host processor 110. The second virtual switch 114 can be capable of receiving packets and classifying them prior to distribution to different hardware schedulers based on a defined arbitration and scheduling scheme. The hardware scheduler 116 receives packets that can be assigned to flow sessions that are scheduled for processing in one or more separate session.

A scheduler 116 can control processing tasks for execution by offload processors 118, including the switching of contexts. In some embodiments, metadata included within data received over memory bus 124 (or metadata derived from such data) can be used to by scheduler 116 the schedule/switch tasks of the offload processors 118. However, command based control of a scheduler via memory bus received commands or flags is also contemplated.

In the particular embodiment of FIG. 1-0, a scheduler 116 can be employed to implement traffic management of incoming packets. Packets from a certain source, relating to a certain traffic class, pertaining to a specific application or flowing to a certain socket are referred to as part of a session flow and are classified using session metadata. Session metadata often serve as the criterion by which packets are prioritized and as such, incoming packets can be reordered based on their session metadata. This reordering of packets can occur in one or more buffers and can modify the traffic shape of these flows. Packets of a session that are reordered based on session metadata can be sent over to specific traffic managed queues that are arbitrated out to output ports using an arbitration circuit (not shown). The arbitration circuit can feed these packet flows to a downstream packet processing/terminating resource directly. Certain embodiments provide for integration of thread and queue management so as to enhance the throughput of downstream resources handling termination of network data through above said threads.

Referring still to FIG. 1-0, data arriving at the scheduler 116 may also be packet data waiting to be terminated at the offload processors 118 or it could be packet data waiting to be processed, modified or switched out. The scheduler 116 can be responsible for segregating incoming packets into corresponding application sessions based on examination of packet data. The scheduler 116 can have circuits for packet inspection and identifying relevant packet characteristics. In some embodiments, a scheduler 116 can offload part of the network stack to free offload processors 118 from overhead incurred from network stack processing. In particular embodiments, a scheduler 116 can carry out any of: TCP/transport offload, encryption/decryption offload, segmentation and reassembly thus allowing offload processors to operate on payloads of network packets directly.

A scheduler 116 can further have the capability to transfer the packets belonging to a session into a particular traffic management queue. A scheduler 116 can to control the scheduling of each of multiple such sessions into a general purpose OS. The stickiness of sessions across a pipeline of stages, including a general purpose OS, can be supported by scheduler 116 optimizing operations carried out at each of the stages. Particular embodiments of such operations are described in more detail below.

While a scheduler 116 have any suitable form, a scheduling circuit which can be used all, or in part, as a scheduler is shown in U.S. Pat. No. 7,760,715, issued to Dalal on Jul. 20, 2010 (hereinafter the '715 patent), and is incorporated herein by reference. The '751 patent discloses a scheduling circuit that takes account of downstream execution resources. The session flows in each of these queues is sent out through an output port to a downstream network element.

A scheduler can employ an arbitration circuit to mediate access of multiple traffic management output queues to available output ports. Each of the output ports may be connected to one of the offload processor cores through a packet buffer. The packet buffer may further include a header pool and a packet body pool. The header pool may only contain the header of packets to be processed by offload processors 118. Sometimes, if the size of the packet to be processed is sufficiently small, the header pool may contain the entire packet. Packets can be transferred over to the header pool or packet body pool depending on the nature of operation carried out at the offload processor 118. For packet processing, overlay, analytics, filtering and such other applications it might be appropriate to transfer only the packet header to the offload processors 118. In this case, depending on the handling of the packet header, the packet body might either be sewn together with a packet header and transferred over an egress interface or dropped. For applications requiring the termination of packets, the entire body of the packet might be transferred. Offload processors can thus receive the packets and execute suitable application session on them.

A scheduler 116 can schedule different sessions on the offload processors 118, acting to coordinate such sessions to reduce the overhead during context switches. A scheduler 116 can arbitrate, not just between outgoing queues or session flows at line rate speeds, but also between terminated sessions at very high speeds. A scheduler 116 can manage the queuing of sessions on offload processors 118 and be responsible for invoking new application sessions on the OS. A scheduler 116 can indicate to the OS that packets for a new session are available based on traffic.

A scheduler 116 can also be informed of the state of the execution resources of offload processors 118, the current session that is run on the execution resource and the memory space allocated to it, as well as the location of the session context in the offload processor cache. A scheduler 116 can thus use the state of the execution resource to carry out traffic management and arbitration decisions.

In the embodiment shown, a scheduler 116 can provide for an integration of thread management on the operating system with traffic management of incoming packets. It can induce a persistence of session flows across a spectrum of components including traffic management queues and processing entities on the offload processors 118. An OS running on an offload processor 118 can allocate execution resources such as processor cycles and memory to a particular queue it is currently handling. The OS can further allocate a thread or a group of threads for that particular queue, so that it can be handled distinctly by the general purpose processing element as a separate entity. Having multiple sessions running on a general purpose (GP) processing resource (e.g., offload processor resource), each handling data from a particular session flow resident in a queue on the scheduler 116, can tightly integrate the scheduler 116 and the GP processing resource. This can bring an element of persistence within session information across the traffic management and scheduler 116 and the GP processing resource.

In some embodiments, an offload processor 118 OS can be modified from a previous OS to reduce the penalty and overhead associated with context switch between resources. This can be further exploited by the hardware scheduler to carry out seamless switching between queues, and consequently their execution as different sessions by the execution resource.

According to particular embodiments, a scheduler 116 can implement traffic management of incoming packets. Packets from a certain source, relating to a certain traffic class, pertaining to a specific application or flowing to a certain socket are referred to as part of a session flow and are classified using session metadata. Session metadata can serve as the criterion by which packets are prioritized and as such, incoming packets are reordered based on their session metadata. This reordering of packets can occur in one or more buffers and can modify the traffic shape of these flows. Packets of a session that are reordered based on session metadata can be sent over to specific traffic managed queues that are arbitrated out to output ports using an arbitration circuit. An arbitration circuit can feed these packet flows to a downstream packet processing and/or terminating resource directly (e.g., offload processor). Certain embodiments provide for integration of thread and queue management so as to enhance the throughput of downstream resources handling termination of network data through above the threads.

In addition to carrying out traffic management, arbitration and scheduling of incoming network packets (and flows), a scheduler 116 can be responsible for enabling minimal overhead context switching between terminated sessions on the OS offload processor 118. Switching between multiple sessions on sessions of the offload processors 118 can makes it possible to terminate multiple sessions at very high speeds. In the embodiment shown, rapid context switching can occur by operation of context module 120. In particular embodiments, a context memory 120 can provide for efficient, low latency context services in a system 100.

In the particular embodiment shown, packets can be transferred over to the scheduler 116 by operation of second switch 114. Scheduler 116 can be responsible both for switching between a session and a new session on the offload processors 118, as well as initiating the saving of context in context memory 120. The context of a session can include, but is not limited to: the state of the processor registers saved in register save area, the instructions in the pipeline being executed, stack pointer and program counter, prefetched instructions and data that are waiting to be executed by the session, data written into the cache recently and any other relevant information that can identify a session executing on the offload processor 118. In particular embodiments, a session context can be identified using the combination of session id, session index in the cache, and starting physical address.

As will be described in more detail with respect to FIG. 1-3, a translation scheme can be used such that contiguous pages of a session in virtual memory are physically contiguous in a cache of an offload processor 118. This contiguous nature of the session in the cache can allow for a bulk read out of the session context into a ‘context snapshot’ for storage in context memory 120, from where it can be retrieved when the operating system (OS) switches processor resources back to the session. The ability to seamlessly fetch session context from a context memory 120 (which can be a low latency memory, and thus be orders of magnitude faster than a main memory of a system) can serve to effectively expand the size of an L2 cache of offload processors 118.

In some embodiments, an OS of system 100 can implement optimizations in its input output memory management unit (IOMMU) (not shown) to allow a translation lookaside buffer (TLB) (or equivalent lookup structure) to distinctly identify contents of each session. Such an arrangement can allow address translations to be identified distinctly during a session switch out and transferred to a page table cache that is external to the TLB. Use of a page table cache can allow for an expansion in the size of the TLB. Also given the fact that contiguous locations in the virtual memory are at contiguous locations in physical memory and in a physically indexed cache, the number of address translations required for identifying a session are significantly reduced.

In the particular embodiment of FIG. 1-0, a system 100 can be well suited to provide out session and packet termination services. In some embodiments, control of network stack processing can be performed by a scheduler 116. Thus, a scheduler 116 can act as a traffic management queue, arbitration circuit and a network stack offload device. The scheduler 116 can be responsible for handling entire session and flow management on behalf of offload processors 118. In such an arrangement, offload processors 118 can be fed with the packets pertaining to a session directly into a buffer, from where it can extract packet data for use. Processing of a network stack can be optimized to avoid switches to a kernel mode to handle network generated interrupts (and execute an interrupt service routine). In this way, a system 100 can be optimized to carry out context switching of sessions seamlessly and with as little overhead as possible.

Referring still to FIG. 1-0, as will be understood, multiple types of conventional input/output busses such as PCI, Fibre Channel can be used in the described system 100. The bus architecture can also be based on relevant JEDEC standards, on DIMM data transfer protocols, on HyperTransport, or any other high speed, low latency interconnection system. Offload processors 118 may include DDR DRAM, RLDRAM, embedded DRAM, next generation stacked memory such as Hybrid Memory Cube (HMC), flash, or other suitable memory, separate logic or bus management chips, programmable units such as field programmable gate arrays (FPGAs), custom designed application specific integrated circuits (ASICs) and an energy efficient, general purpose processor such as those based on ARM, ARC, Tensilica, MIPS, Strong/ARM, or RISC architectures. Host processors 110 can be a general purpose processor, including those based on Intel or AMD x86 architecture, Intel Itanium architecture, MIPS architecture, SPARC architecture or the like.

As will also be understood, conventional systems executing processing like that performed by the system of FIG. 1-0 can be implemented on multiple threads running on multiple processing cores. Such parallelization of tasks into multiple thread contexts can provide for increased throughput. Processor architectures such as MIPS may include deep instruction pipelines to improve the number of instructions per cycle. Further, the ability to run a multi-threaded programming environment results in enhanced usage of existing processor resources. To further increase parallel execution on the hardware, processor architectures may include multiple processor cores. Multi-core architectures comprising the same type of cores, referred to as homogeneous core architectures, provide higher instruction throughput by parallelizing threads or processes across multiple cores. However, in such homogeneous core architectures, the shared resources, such as memory, are amortized over a small number of processors. In still other embodiments, multiple offload or host processors can reside on modules connected to individual rack units or blades that in turn reside on racks or individual servers. These can be further grouped into clusters and datacenters, which can be spatially located in the same building, in the same city, or even in different countries. Any grouping level can be connected to each other, and/or connected to public or private cloud internets.

In such conventional approaches, memory and I/O accesses can incur a high amount of processor overhead. Further, as noted herein, context switches in conventional general purpose processing units can be computationally intensive. It is therefore desirable to reduce context switch overhead in a networked computing resource handling a plurality of networked applications in order to increase processor throughput. Conventional server loads can require complex transport, high memory bandwidth, extreme amounts of data bandwidth (randomly accessed, parallelized, and highly available), but often with light touch processing: HTML, video, packet-level services, security, and analytics. Further, idle processors still consume more than 50% of their peak power consumption.

In contrast, in an embodiment such as that shown in FIG. 1-0, or an equivalent, complex transport, data bandwidth intensive, frequent random access oriented, “light touch” processing loads can be handled behind a socket abstraction created on multiple offload processor 118 cores. At the same time, “heavy touch”, computing intensive loads can be handled by a socket abstraction on a host processor 110 core (e.g., x86 processor cores). Such software sockets can allow for a natural partitioning of these loads between light touch (e.g., ARM) and heavy touch (e.g., x86) processor cores. By usage of new application level sockets, according to embodiments, server loads can be broken up across the offload processors 118 and the host processor(s) 110.

To better understand operations of the embodiments disclosed herein, conventional cache schemes are described with reference to FIGS. 1-1 and 1-2. Modern operating systems that implement virtual memory are responsible for the allocation of both virtual and physical memory for processes, resulting in virtual to physical translations that occur when a process executes and accesses virtually addressed memory. In the management of memory for a process, there is typically no coordination between the allocation of a virtual address range and the corresponding physical addresses that will be mapped by the virtual addresses. This lack of coordination can affect both processor cache overhead and effectiveness when a process is executing.

In conventional systems, a processor allocates memory pages that are contiguous in virtual memory for each process that is executing. The processor also allocates pages in physical memory, which are not necessarily contiguous. A translation scheme is established between the two schemes of addressing to ensure that the abstraction of virtual memory is correctly supported by physical memory pages. A processor can employ cache blocks that are resident close to the processor to meet the immediate data processing needs. Conventional caches can be arranged in a hierarchy. Level One (L1) caches are closest to the processor, followed by L2, L3, and so on. L2 acts a backup to L1 and so on. For caches that are indexed by a part of the process's physical addresses, the lack of correlation between the allocation of virtual and physical memory for a range of addresses beyond the size of a memory management unit (MMU) page results in haphazard and inefficient effects in the processor caches. This increases cache overheads and delay is introduced during a context switch operation.

In physically addressed caches, the cache entry for the next page in the virtual memory may not correspond to the next contiguous page in the cache—thus degrading the overall performance that can be achieved. For example, in FIG. 1-1, contiguous pages in virtual memory 130 (Pages 1 and 2 of Process 1) collide in the cache as their physical addresses in physical memory 132 index to the same location of the physically indexed cache 134 (of the processor). That is, processor cache (i.e., 134) is physically indexed, and the addresses of the pages in the physical memory 132 index to the same page in the processor cache. Furthermore, when the effects of multiple processes accessing a shared cache are considered, there is typically a lack of consideration of overall cache performance when the OS allocates physical memory to processes. This lack of consideration results in different processes thrashing in the cache across context switches (e.g., Process 1 and Process 2 in FIG. 1-1), which can unnecessarily displace each other's lines, which can result in an indeterminate number of cache miss/fills upon resuming a process, or an increased number of line writebacks across context switch.

As described with reference to FIG. 1-2, in other conventional arrangements, processor caches can alternatively be indexed by a part of the process's virtual addresses. Virtually indexed caches are accessed by using a section of the bits of the virtual address of the processor. Pages that are contiguous in virtual memory 130 will be contiguous in virtually indexed caches 136 as seen in FIG. 1-2. As long as processor caches are virtually indexed, no attention needs to be paid to coordinating the allocation of physical memory 132 with the allocation of virtual addresses. As programs sweep through virtual address ranges, they will enjoy the benefits of spatial locality in the processor cache. Such set-associative caches can have several entries corresponding to an index. A given page which maps onto the given cache index can be anywhere in that particular set. Given that there are several positions available for a cache entry, the problems that caused thrashing in the cache across context switches (i.e., as shown in FIG. 1-1) are alleviated to a certain extent with set-associative caches, as the processor can afford to keep used entries in the cache to the longest extent possible. For this, caches employ a least recently used algorithm. This results in mitigation of some of the problems associated with a virtual addressing scheme followed by an operating system, but places constraints on the size of the cache. Consequently, bigger, multi-way associative caches can be required to ensure that recently used entries are not invalidated/flushed out. The comparator circuitry for a multi-way set associative cache can be complex to accommodate for parallel comparison, which increases the circuit level complexity associated with the cache.

A cache control scheme known as “page coloring” has been used by some conventional operating systems to deal with the problem of cache-misses due to a virtual addressing scheme. If the processor cache was physically indexed, the operating system was constrained to look for physical memory locations that would not index to locations in the cache of the same color. Under such a cache control scheme, an operating system would have to assess, for every virtual address, those pages in the physical memory that are allowable based on the index they hash to in the physically indexed cache. Several physical addresses are disallowed as the indices derived might be of the same color. So, for physically indexed caches, every page in the virtual memory would be colored to identify its corresponding cache location and determine if the next page is allocated to a physical memory and thus a cache location of the same color or not. This process would be repeated for every page, which can be a cumbersome operation. While it improves cache efficiency, page coloring increases the overhead on the memory management and translation unit as colors of every page would have to be identified to prevent recently used pages from being overwritten. The level of complexity of the operating system increases correspondingly, as it needs to maintain an indicator of the color of the previous virtual memory page in the cache.

The problem with a virtually indexed cache is that despite the fact that the cache access latencies are higher, there is the pervasive problem of aliasing. In the case of aliasing, multiple virtual addresses (with different indices) mapping to the same page in the physical memory are at different locations in the cache (due to the different indices). Page coloring allows the virtual pages and physical pages to have the same color and therefore occupy the same set in the cache. Page coloring makes aliases to share the same superset bits and index to the same lines in the cache. This removes the problem of aliasing. Page coloring also imposes constraints on memory allocation. When a new physical page is allocated on a page fault, a memory management algorithm must pick a page with the same color as the virtual color from the free list. Because systems allocate virtual space systematically, the pages of different programs tend to have the same colors, and thus some physical colors may be more frequent than others. Thus page coloring may impact the page fault rate. Moreover, the predominance of some physical colors may create mapping conflicts between programs in a second-level cache accessed with physical addresses. Thus, a processor is faced with a very big problem with the conventional page coloring scheme just described. Each of the virtual pages could be occupying different pages in the physical memory such that they occupy different cache colors, but the processor would need to store the address translation of each and every page. Given that a process could be sufficiently large, and each process would comprise of several virtual pages, the page coloring algorithm could become very complex. This would also complicate it at the TLB end, as it would need to identify for each page of the processor's virtual memory, the equivalent physical address. As context switches tend to invalidate the TLB entries, the processor would need to carry out page walks and fill the TLB entries, and this would further add indeterminism and latency to what is a routine context switch.

In this way, in commonly available conventional operating systems, context switches result in collisions in the cache as well as TLB misses when a process/thread is resumed. When the process/thread resumes, there are an indeterminate number of instruction and data cache misses as the thread's working set is reloaded back into the cache (i.e., as the thread resumes in user space and executes instructions, the instructions will typically have to be loaded into the cache, along with the application data). Upon switch-in (i.e., resumption of process/thread), the TLB mappings may be completely or partially invalidated, with the base of the new thread's page tables written to a register reserved for that purpose. As the thread executes, the TLB misses will result in page table walks (either by hardware or software) which result in TLB fills. Each of these TLB misses has its own hardware costs, including pipeline stall due to an exception (e.g., the overhead created by memory accesses when performing a page table walk, along with the associated cache misses/memory loads if the page tables are not in the cache). These costs are dependent upon what took place in the processor between successive runs of a process and are therefore not fixed costs. Furthermore, these extra latencies add to the cost of a context switch and detract from the effective execution of a process. As will be appreciated, such foregoing described cache control methods are non-deterministic with respect to processing time, memory requirements, or other operating system controlled resources, reducing overall efficiency of system operation.

FIG. 1-3 shows a cache control system according to an embodiment. In the cache control system, session contents can be contiguous in a physically indexed cache 134′. The described embodiment can use a translation scheme such that contiguous pages of a session in virtual memory 130 are physically contiguous in the physically indexed cache 134. In contrast to the foregoing described non-deterministic cache control schemes, at least the duration of a context switch operation can be deterministic. In the described embodiment, replacing the context of a previous process by the context of a new process involves transferring the new process context from an external low latency memory such as provided by context memory 120 of FIG. 1-0. In the process of context switching, access of a main memory of a system can be avoided (where such accesses can be delay intensive). The process context is prefetched from the context memory 120 (which can be a low latency memory). If needed for another context switch, process context can be saved once again to the context memory 120. In this way, deterministic context switching is achieved, as a context switch operation can be defined in terms of the number of cycles and the operations needed to be carried out. Further, use of a low latency memory to store context data can provide for rapid context switching.

FIGS. 2-0 to 2-5 describe aspects of hardware embodiments of a module that can include context switching as described herein. In particular embodiments, such processing modules can include DIMM mountable modules.

FIG. 2-0 is a block diagram of a processing module 200 according to one embodiment. A processing module 200 can include a physical connector 202, a memory interface 204, arbiter logic 206, offload processor(s) 208, local memory 210, and control logic 212. A connector 202 can provide a physical connection to system memory bus. This is in contrast to a host processor which can access a system memory bus via a memory controller, or the like. In very particular embodiments, a connector 202 can be compatible with a dual in-line memory module (DIMM) slot of a computing system. Accordingly, a system including multiple DIMM slots can be populated with one or more processing modules 200, or a mix of processing modules and DIMM modules.

A memory interface 204 can detect data transfers on a system memory bus, and in appropriate cases, enable write data to be stored in the processing module 200 and/or read data to be read out from the processing module 200. Such data transfers can include the receipt of packet data having a particular network identifier. In some embodiments, a memory interface 204 can be a slave interface, thus data transfers are controlled by a master device separate from the processing module 200. In very particular embodiments, a memory interface 204 can be a direct memory access (DMA) slave, to accommodate DMA transfers over a system memory bus initiated by a DMA master. In some embodiments, a DMA master can be a device different from a host processor. In such configurations, processing module 200 can receive data for processing (e.g., DMA write), and transfer processed data out (e.g., DMA read) without consuming host processor resources.

Arbiter logic 206 can arbitrate between conflicting accesses of data within processing module 200. In some embodiments, arbiter logic 206 can arbitrate between accesses by offload processor 208 and accesses external to the processor module 200. It is understood that a processing module 200 can include multiple locations that are operated on at the same time. It is understood that accesses arbitrated by arbiter logic 206 can include accesses to physical system memory space occupied by the processor module 200, as well as accesses to other resources (e.g., cache memory of offload or host processor). Accordingly, arbitration rules for arbiter logic 206 can vary according to application. In some embodiments, such arbitration rules are fixed for a given processor module 200. In such cases, different applications can be accommodated by switching out different processing modules. However, in alternate embodiments, such arbitration rules can be configurable.

Offload processor 208 can include one or more processors that can operate on data transferred over the system memory bus. In some embodiments, offload processors can run a general operating system or server applications such as Apache (as but one very particular example), enabling processor contexts to be saved and retrieved. Computing tasks executed by offload processor 208 can be handled by the hardware scheduler. Offload processors 208 can operate on data buffered in the processor module 200. In addition or alternatively, offload processors 208 can access data stored elsewhere in a system memory space. In some embodiments, offload processors 208 can include a cache memory configured to store context information. An offload processor 208 can include multiple cores or one core.

A processor module 200 can be included in a system having a host processor (not shown). In some embodiments, offload processors 208 can be a different type of processor as compared to the host processor. In particular embodiments, offload processors 208 can consume less power and/or have less computing power than a host processor. In very particular embodiments, offload processors 208 can be “wimpy” core processors, while a host processor can be a “brawny” core processor. However, in alternate embodiments, offload processors 208 can have equivalent computing power to any host processor. In very particular embodiments, a host processor can be an x86 type processor, while an offload processor 208 can include an ARM, ARC, Tensilica, MIPS, Strong/ARM, or RISC type processor, as but a few examples.

Local memory 210 can be connected to offload processor 208 to enable the storing of context information. Accordingly, an offload processor 208 can store current context information, and then switch to a new computing task, then subsequently retrieve the context information to resume the prior task. In very particular embodiments, local memory 210 can be a low latency memory with respect to other memories in a system. In some embodiments, storing of context information can include copying an offload processor 208 cache.

In some embodiments, a same space within local memory 210 is accessible by multiple offload processors 208 of the same type. In this way, a context stored by one offload processor can be resumed by a different offload processor.

Control logic 212 can control processing tasks executed by offload processor(s). In some embodiments, control logic 212 can be considered a hardware scheduler that can be conceptualized as including a data evaluator 214, scheduler 216 and a switch controller 218. A data evaluator 214 can extract “metadata” from write data transferred over a system memory bus. “Metadata”, as used herein, can be any information embedded at one or more predetermined locations of a block of write data that indicates processing to be performed on all or a portion of the block of write data and/or indicate a particular task/process to which the data belongs (e.g., classification data). In some embodiments, metadata can be data that indicates a higher level organization for the block of write data. As but one very particular embodiment, metadata can be header information of one or more network packets (which may or may not be encapsulated within a higher layer packet structure).

A scheduler 216 (e.g., a hardware scheduler) can order computing tasks for offload processor(s) 208. In some embodiments, scheduler 216 can generate a schedule that is continually updated as write data for processing is received. In very particular embodiments, a scheduler 216 can generate such a schedule based on the ability to switch contexts of offload processor(s) 208. In this way, on-module computing priorities can be adjusted on the fly. In very particular embodiments, a scheduler 216 can assign a portion of physical address space (e.g., memory locations within local memory 210) to an offload processor 208, according to computing tasks. The offload processor 208 can then switch between such different spaces, saving context information prior to each switch, and subsequently restoring context information when returning to the memory space.

Switch controller 218 can control computing operations of offload processor(s) 208. In particular embodiments, according to scheduler 216, switch controller 218 can order offload processor(s) 208 to switch contexts. It is understood that a context switch operation can be an “atomic” operation, executed in response to a single command from switch controller 218. In addition or alternatively, a switch controller 218 can issue an instruction set that stores current context information, recalls context information, etc.

In some embodiments, processor module 200 can include a buffer memory (not shown). A buffer memory can store received write data on board the processor module. A buffer memory can be implemented on an entirely different set of memory devices, or can be a memory embedded with logic and/or the offload processor. In the latter case, arbiter logic 206 can arbitrate access to the buffer memory. In some embodiments, a buffer memory can correspond to a portion of a system physical memory space. The remaining portion of the system memory space can correspond to other like processor modules and/or memory modules connected to the same system memory bus. In some embodiments buffer memory can be different than local memory 210. For example, buffer memory can have a slower access time than local memory 210. However, in other embodiments, buffer memory and local memory can be implemented with like memory devices.

In very particular embodiments, write data for processing can have an expected maximum flow rate. A processor module 200 can be configured to operate on such data at, or faster than, such a flow rate. In this way, a master device (not shown) can write data to a processor module without danger of overwriting data “in process”.

The various computing elements of a processor module 200 can be implemented as one or more integrated circuit devices (ICs). It is understood that the various components shown in FIG. 2-0 can be formed in the same or different ICs. For example, control logic 212, memory interface 214, and/or arbiter logic 206 can be implemented on one or more logic ICs, while offload processor(s) 208 and local memory 210 are separate ICs. Logic ICs can be fixed logic (e.g., application specific ICs), programmable logic (e.g., field programmable gate arrays, FPGAs), or combinations thereof.

Advantageously, the foregoing hardware and systems can provide improved computational performance as compared to traditional computing systems. Conventional systems, including those based on x86 processors, are often ill-equipped to handle such high volume applications. Even idling, x86 processors use a significant amount of power, and near continuous operation for high bandwidth packet analysis or other high volume processing tasks makes the processor energy costs one of the dominant price factors.

In addition, conventional systems can have issues with the high cost of context switching wherein a host processor is required to execute instructions which can include switching from one thread to another. Such a switch can require storing and recalling the context for the thread. If such context data is resident in a host cache memory, such a context switch can occur relatively quickly. However, if such context data is no longer in cache memory (i.e., a cache miss), the data must be recalled from system memory, which can incur a multi-cycle latency. Continuous cache misses during context switching can adversely impact system performance.

FIG. 2-1 shows a processor module 200-1 according to one very particular embodiment which is capable of reducing issues associated with high volume processing or context switching associated with many conventional server systems. A processor module 200-1 can include ICs 220-0/1 mounted to a printed circuit board (PCB) type substrate 222. PCB type substrate 222 can include in-line module connector 202, which in one very particular embodiment, can be a DIMM compatible connector. IC 220-0 can be a system-on-chip (SoC) type device, integrating multiple functions. In the very particular embodiment shown, an IC 220-0 can include embedded processor(s), logic and memory. Such embedded processor(s) can be offload processor(s) 208 as described herein, or equivalents. Such logic can be any of controller logic 212, memory interface 204 and/or arbiter logic 206, as described herein, or equivalents. Such memory can be any of local memory 210, cache memory for offload processor(s) 208, or buffer memory, as described herein, or equivalents. Logic IC 220-1 can provide logic functions not included IC 220-0.

FIG. 2-2 shows a processor module 200-2 according to another very particular embodiment. A processor module 200-2 can include ICs 220-2, -3, -4, -5 mounted to a PCB type substrate 222, like that of FIG. 2-1. However, unlike FIG. 2-1, processor module functions are distributed among single purpose type ICs. IC 220-2 can be a processor IC, which can be an offload processor 208. IC 220-3 can be a memory IC which can include local memory 210, buffer memory, or combinations thereof. IC 220-4 can be a logic IC which can include control logic 212, and in one very particular embodiment, can be an FPGA. IC 220-5 can be another logic IC which can include memory interface 204 and arbiter logic 206, and in one very particular embodiment, can also be an FPGA.

It is understood that FIGS. 2-1/2 represent but two of various implementations. The various functions of a processor module can be distributed over any suitable number of ICs, including a single SoC type IC.

FIG. 2-3 shows an opposing side of a processor module 200-1 or 200-2 according to a very particular embodiment. Processor module 200-3 can include a number of memory ICs, one shown as 220-6, mounted to a PCB type substrate 222, like that of FIG. 2-1. It is understood that various processing and logic components can be mounted on an opposing side to that shown. A memory IC 220-6 can be configured to represent a portion of the physical memory space of a system. Memory ICs 220-6 can perform any or all of the following functions: operate independently of other processor module components, providing system memory accessed in a conventional fashion; serve as buffer memory, storing write data that can be processed with other processor module components, or serve as local memory for storing processor context information.

FIG. 2-4 shows a conventional DIMM module (i.e., it serves only a memory function) that can populate a memory bus along with processor modules as described herein, or equivalents.

FIG. 2-5 shows a system 230 according to one embodiment. A system 230 can include a system memory bus 228 accessible via multiple in-line module slots (one shown as 226). According to embodiments, any or all of the slots 226 can be occupied by a processor module 200 as described herein, or an equivalent. In the event all slots 226 are not occupied by a processor module 200, available slots can be occupied by conventional in-line memory modules 224. In a very particular embodiment, slots 226 can be DIMM slots.

In some embodiments, a processor module 200 can occupy one slot. However, in other embodiments, a processor module can occupy multiple slots.

In some embodiments, a system memory bus 228 can be further interfaced with one or more host processors and/or input/output device (not shown).

Having described processor modules according to various embodiments, operations of an offload processor module capable of interfacing with server or similar system via a memory bus and according to a particular embodiment will now be described.

FIG. 3 shows a system 301 that can execute context switches in offload processors according to an embodiment. In the example shown, a system 301 can transport packet data to one or more computational units (one shown as 300) located on a module, which in particular embodiments, can include a connector compatible with an existing memory module. In some embodiments, a computational unit 300 can include a processor module as described in embodiments herein, or an equivalent. A computational unit 300 can be capable of intercepting or otherwise accessing packets sent over a memory bus 316 and carrying out processing on such packets, including but not limited to termination or metadata processing. A system memory bus 316 can be a system memory bus like those described herein, or equivalents (e.g., 228).

Referring still to FIG. 3, a system 301 can include an I/O device 302 which can receive packet or other I/O data from an external source. In some embodiments I/O device 302 can include physical or virtual functions generated by the physical device to receive a packet or other I/O data from the network or another computer or virtual machine. In the very particular embodiment shown, an I/O device 302 can include a network interface card (NIC) having input buffer 302a (e.g., DMA ring buffer) and an I/O virtualization function 302b.

According to embodiments, an I/O device 302 can write a descriptor including details of the necessary memory operation for the packet (i.e. read/write, source/destination). Such a descriptor can be assigned a virtual memory location (e.g., by an operating system of the system 301). I/O device 302 then communicates with an input output memory management unit (IOMMU) 304 which can translate virtual addresses to corresponding physical addresses with an IOMMU function 304b. In the particular embodiment shown, a translation look-aside buffer (TLB) 304a can be used for such translation. Virtual function reads or writes data between I/O device and system memory locations can then be executed with a direct memory transfer (e.g., DMA) via a memory controller 306b of the system 301. An I/O device 302 can be connected to IOMMU 304 by a host bus 312. In one very particular embodiment, a host bus 312 can be a peripheral interconnect (PCI) type bus. IOMMU 304 can be connected to a host processing section 306 at a central processing unit I/O (CPUI0) 306a. In the embodiment shown, such a connection 314 can support a HyperTransport (HT) protocol.

In the embodiment shown, a host processing section 306 can include the CPUIO 306a, memory controller 306b, processing core 306c and corresponding provisioning agent 306d.

In particular embodiments, a computational unit 300 can interface with the system bus 316 via standard in-line module connection, which in very particular embodiments can include a DIMM type slot. In the embodiment shown, a memory bus 316 can be a DDR3 type memory bus. Alternate embodiments can include any suitable system memory bus. Packet data can be sent by memory controller 306b via memory bus 316 to a DMA slave interface 310a. DMA slave interface 310a can be adapted to receive encapsulated read/write instructions from a DMA write over the memory bus 316.

A hardware scheduler (308b/c/d/e/h) can perform traffic management on incoming packets by categorizing them according to flow using session metadata. Packets can be queued for output in an onboard memory (310b/308a/308m) based on session priority. When the hardware scheduler determines that a packet for a particular session is ready to be processed by the offload processor 308i, the onboard memory is signaled for a context switch to that session. Utilizing this method of prioritization, context switching overhead can be reduced, as compared to conventional approaches. That is, a hardware scheduler can handle context switching decisions and thus optimize the performance of the downstream resource (e.g., offload processor 308i).

As noted above, in very particular embodiments, an offload processor 308i can be a “wimpy core” type processor. According to some embodiments, a host processor 306c can be a “brawny core” type processor (e.g., an x86 or any other processor capable of handling “heavy touch” computational operations). While an I/O device 302 can be configured to trigger host processor interrupts in response to incoming packets, according to embodiments, such interrupts can be disabled, thereby reducing processing overhead for the host processor 306c. In some very particular embodiments, an offload processor 308i can include an ARM, ARC, Tensilica, MIPS, Strong/ARM or any other processor capable of handling “light touch” operations. Preferably, an offload processor can run a general purpose operating system for executing a plurality of sessions, which can be optimized to work in conjunction with the hardware scheduler in order to reduce context switching overhead.

Referring still to FIG. 3, in operation, a system 301 can receive packets from an external network over a network interface. The packets are destined for either a host processor 306c or an offload processor 308i based on the classification logic and schematics employed by I/O device 302. In particular embodiments, I/O device 302 can operate as a virtualized NIC, with packets for a particular logical network or to a certain virtual MAC (VMAC) address can be directed into separate queues and sent over to the destination logical entity. Such an arrangement can transfer packets to different entities. In some embodiments, each such entity can have a virtual driver, a virtual device model that it uses to communicate with connected virtual network.

According to embodiments, multiple devices can be used to redirect traffic to specific memory addresses. So, each of the network devices operates as if it is transferring the packets to the memory location of a logical entity. However, in reality, such packets are transferred to memory addresses where they can be handled by one or more offload processors (e.g., 308i). In particular embodiments such transfers are to physical memory addresses, thus logical entities can be removed from the processing, and a host processor can be free from such packet handling.

Accordingly, embodiments can be conceptualized as providing a memory “black box” to which specific network data can be fed. Such a memory black box can handle the data (e.g., process it) and respond back when such data is requested.

Referring still to FIG. 3, according to some embodiments, I/O device 302 can receive data packets from a network or from a computing device. The data packets can have certain characteristics, including transport protocol number, source and destination port numbers, source and destination IP addresses, for example. The data packets can further have metadata that is processed (308d) that helps in their classification and management.

I/O device 302 can include, but is not limited to, peripheral component interconnect (PCI) and/or PCI express (PCIe) devices connecting with a host motherboard via PCI or PCIe bus (e.g., 312). Examples of I/O devices include a network interface controller (NIC), a host bus adapter, a converged network adapter, an ATM network interface, etc.

In order to provide for an abstraction scheme that allows multiple logical entities to access the same I/O device 302, the I/O device may be virtualized to provide for multiple virtual devices each of which can perform some of the functions of the physical I/O device. The IO virtualization program (e.g., 302b) according to an embodiment, can redirect traffic to different memory locations (and thus to different offload processors attached to modules on a memory bus). To achieve this, an I/O device 302 (e.g., a network card) may be partitioned into several function parts; including controlling function (CF) supporting input/output virtualization (IOV) architecture (e.g., single-root IOV) and multiple virtual function (VF) interfaces. Each virtual function interface may be provided with resources during runtime for dedicated usage. Examples of the CF and VF may include the physical function and virtual functions under schemes such as Single Root I/O Virtualization or Multi-Root I/O Virtualization architecture. The CF acts as the physical resources that sets up and manages virtual resources. The CF is also capable of acting as a full-fledged 10 device. The VF is responsible for providing an abstraction of a virtual device for communication with multiple logical entities/multiple memory regions.

The operating system/the hypervisor/any of the virtual machines/user code running on a host processor 306c may be loaded with a device model, a VF driver and a driver for a CF. The device model may be used to create an emulation of a physical device for the host processor 306c to recognize each of the multiple VFs that are created. The device model may be replicated multiple times to give the impression to a VF driver (a driver that interacts with a virtual IO device) that it is interacting with a physical device of a particular type.

For example, a certain device module may be used to emulate a network adapter such as the Intel® Ethernet Converged Network Adapter(CNA) X540-T2, so that the I/O device 302 believes it is interacting with such an adapter. In such a case, each of the virtual functions may have the capability to support the functions of the above said CNA, i.e., each of the Physical Functions should be able to support such functionality. The device model and the VF driver can be run in either privileged or non-privileged mode. In some embodiments, there is no restriction with regard to who hosts/runs the code corresponding to the device model and the VF driver. The code, however, has the capability to create multiple copies of device model and VF driver so as to enable multiple copies of said I/O interface to be created.

An application or provisioning agent 306d, as part of an application/user level code running in a kernel, may create a virtual I/O address space for each VF, during runtime and allocate part of the physical address space to it. For example, if an application handling the VF driver instructs it to read or write packets from or to memory addresses 0xaaaa to 0xffff, the device driver may write I/O descriptors into a descriptor queue with a head and tail pointer that are changed dynamically as queue entries are filled. The data structure may be of another type as well, including but not limited to a ring structure 302a or hash table.

The VF can read from or write data to the address location pointed to by the driver. Further, on completing the transfer of data to the address space allocated to the driver, interrupts, which are usually triggered to the host processor to handle said network packets, can be disabled. Allocating a specific I/O space to a device can include allocating said IO space a specific physical memory space occupied.

In another embodiment, the descriptor may comprise only a write operation, if the descriptor is associated with a specific data structure for handling incoming packets. Further, the descriptor for each of the entries in the incoming data structure may be constant so as to redirect all data write to a specific memory location. In an alternate embodiment, the descriptor for consecutive entries may point to consecutive entries in memory so as to direct incoming packets to consecutive memory locations.

Alternatively, said operating system may create a defined physical address space for an application supporting the VF drivers and allocate a virtual memory address space to the application or provisioning agent 306d, thereby creating a mapping for each virtual function between said virtual address and a physical address space. Said mapping between virtual memory address space and physical memory space may be stored in IOMMU tables (e.g., a TLB 304a). The application performing memory reads or writes may supply virtual addresses to say virtual function, and the host processor OS may allocate a specific part of the physical memory location to such an application.

Alternatively, VF may be configured to generate requests such as read and write which may be part of a direct memory access (DMA) read or write operation, for example. The virtual addresses is be translated by the IOMMU 304 to their corresponding physical addresses and the physical addresses may be provided to the memory controller for access. That is, the IOMMU 304 may modify the memory requests sourced by the I/O devices to change the virtual address in the request to a physical address, and the memory request may be forwarded to the memory controller for memory access. The memory request may be forwarded over a bus 314 that supports a protocol such as HyperTransport 314. The VF may in such cases carry out a direct memory access by supplying the virtual memory address to the IOMMU 304.

Alternatively, said application may directly code the physical address into the VF descriptors if the VF allows for it. If the VF cannot support physical addresses of the form used by the host processor 306c, an aperture with a hardware size supported by the VF device may be coded into the descriptor so that the VF is informed of the target hardware address of the device. Data that is transferred to an aperture may be mapped by a translation table to a defined physical address space in the system memory. The DMA operations may be initiated by software executed by the processors, programming the I/O devices directly or indirectly to perform the DMA operations.

Referring still to FIG. 3, in particular embodiments, parts of computational unit 300 can be implemented with one or more FPGAs. In the system of FIG. 3, computational unit 300 can include FPGA 310 in which can be formed a DMA slave device module 310a and arbiter 310f. A DMA slave module 310a can be any device suitable for attachment to a memory bus 316 that can respond to DMA read/write requests. In alternate embodiments, a DMA slave module 310a can be another interface capable of block data transfers over memory bus 316. The DMA slave module 310a can be capable of receiving data from a DMA controller (when it performs a read from a ‘memory’ or from a peripheral) or transferring data to a DMA controller (when it performs a write instruction on the DMA slave module 310a). The DMA slave module 310a may be adapted to receive DMA read and write instructions encapsulated over a memory bus, (e.g., in the form of a DDR data transmission, such as a packet or data burst), or any other format that can be sent over the corresponding memory bus.

A DMA slave module 310a can reconstruct the DMA read/write instruction from the memory R/W packet. The DMA slave module 310a may be adapted to respond to these instructions in the form of data reads/data writes to the DMA master, which could either be housed in a peripheral device, in the case of a PCIe bus, or a system DMA controller in the case of an ISA bus.

I/O data that is received by the DMA device 310a can then queued for arbitration. Arbitration can include the process of scheduling packets of different flows, such that they are provided access to available bandwidth based on a number of parameters. In general, an arbiter 310f provides resource access to one or more requestors. If multiple requestors request access, an arbiter 310f can determine which requestor becomes the accessor and then passes data from the accessor to the resource interface, and the downstream resource can begin execution on the data. After the data has been completely transferred to a resource, and the resource has competed execution, the arbiter 310f can transfer control to a different requestor and this cycle repeats for all available requestors. In the embodiment of FIG. 3 arbiter 310f can notify other portions of computational unit 300 (e.g., 308) of incoming data.

Alternatively, a computation unit 300 can utilize an arbitration scheme shown in U.S. Pat. No. 7,813,283, issued to Dalal on Oct. 12, 2010, the contents of which are incorporated herein by reference. Other suitable arbitration schemes known in art could be implemented in embodiments herein. Alternatively, the arbitration scheme of the current invention might be implemented using an OpenFlow switch and an OpenFlow controller.

In the very particular embodiment of FIG. 3, computational unit 300 can further include notify/prefetch circuits 310c which can prefetch data stored in a buffer memory 310b in response to DMA slave module 310a, and as arbitrated by arbiter 310f. Further, arbiter 310f can access other portions of the computational unit 300 via a memory mapped I/O ingress path 310e and egress path 310g.

Referring to FIG. 3, a hardware scheduler can include a scheduling circuit 308b/n to implement traffic management of incoming packets. Packets from a certain source, relating to a certain traffic class, pertaining to a specific application or flowing to a certain socket are referred to as part of a session flow and are classified using session metadata. Such classification can be performed by classifier 308e.

In some embodiments, session metadata 308d can serve as the criterion by which packets are prioritized and scheduled and as such, incoming packets can be reordered based on their session metadata. This reordering of packets can occur in one or more buffers and can modify the traffic shape of these flows. The scheduling discipline chosen for this prioritization, or traffic management (TM), can affect the traffic shape of flows and micro-flows through delay (buffering), bursting of traffic (buffering and bursting), smoothing of traffic (buffering and rate-limiting flows), dropping traffic (choosing data to discard so as to avoid exhausting the buffer), delay jitter (temporally shifting cells of a flow by different amounts) and by not admitting a connection (e.g., cannot simultaneously guarantee existing service level agreements (SLAs) with an additional flow's SLA).

According to embodiments, computational unit 300 can serve as part of a switch fabric, and provide traffic management with depth-limited output queues, the access to which is arbitrated by a scheduling circuit 308b/n. Such output queues are managed using a scheduling discipline to provide traffic management for incoming flows. The session flows queued in each of these queues can be sent out through an output port to a downstream network element.

It is noted that conventional traffic management do not take into account the handling and management of data by downstream elements except for meeting the SLA agreements it already has with said downstream elements.

In contrast, according to embodiments a scheduler circuit 308b/n can allocate a priority to each of the output queues and carry out reordering of incoming packets to maintain persistence of session flows in these queues. A scheduler circuit 308b/n can be used to control the scheduling of each of these persistent sessions into a general purpose operating system (OS) 308j, executed on an offload processor 308i. Packets of a particular session flow, as defined above, can belong to a particular queue. The scheduler circuit 308b/n may control the prioritization of these queues such that they are arbitrated for handling by a general purpose (GP) processing resource (e.g., offload processor 308i) located downstream. An OS 308j running on a downstream processor 308i can allocate execution resources such as processor cycles and memory to a particular queue it is currently handling. The OS 308j may further allocate a thread or a group of threads for that particular queue, so that it is handled distinctly by the general purpose processing element 308i as a separate entity. The fact that there can be multiple sessions running on a GP processing resource, each handling data from a particular session flow resident in a queue established by the scheduler circuit, tightly integrates the scheduler and the downstream resource (e.g., 308i). This can bring about persistence of session information across the traffic management and scheduling circuit and the general purpose processing resource 308i.

Dedicated computing resources (e.g., 308i), memory space and session context information for each of the sessions can provide a way of handling, processing and/or terminating each of the session flows at the general purpose processor 308i. The scheduler circuit 308b/n can exploit this functionality of the execution resource to queue session flows for scheduling downstream. The scheduler circuit 308b/n can be informed of the state of the execution resource(s) (e.g., 308i), the current session that is run on the execution resource; the memory space allocated to it, the location of the session context in the processor cache.

According to embodiments, a scheduler circuit 308b/n can further include switching circuits to change execution resources from one state to another. The scheduler circuit 308b/n can use such a capability to arbitrate between the queues that are ready to be switched into the downstream execution resource. Further, the downstream execution resource can be optimized to reduce the penalty and overhead associated with context switch between resources. This is further exploited by the scheduler circuit 308b/n to carry out seamless switching between queues, and consequently their execution as different sessions by the execution resource.

According to embodiments, a scheduler circuit 308b/n can schedule different sessions on a downstream processing resource, wherein the two are operated in coordination to reduce the overhead during context switches. An important factor in decreasing the latency of services and engineering computational availability can be hardware context switching synchronized with network queuing. In embodiments, when a queue is selected by a traffic manager, a pipeline coordinates swapping in of the cache (e.g., L2 cache) of the corresponding resource (e.g., 308i) and transfers the reassembled I/O data into the memory space of the executing process. In certain cases, no packets are pending in the queue, but computation is still pending to service previous packets. Once this process makes a memory reference outside of the data swapped, the scheduler circuit (308b/n) can enable queued data from an I/O device 302 to continue scheduling the thread.

In some embodiments, to provide fair queuing to a process not having data, a maximum context size can be assumed as data processed. In this way, a queue can be provisioned as the greater of computational resource and network bandwidth resource. As but one very particular example, a computation resource can be an ARM A9 processor running at 800 MHz, while a network bandwidth can be 3 Gbps of bandwidth. Given the lopsided nature of this ratio, embodiments can utilize computation having many parallel sessions (such that the hardware's prefetching of session-specific data offloads a large portion of the host processor load) and having minimal general purpose processing of data.

Accordingly, in some embodiments, a scheduler circuit 308b/n can be conceptualized as arbitrating, not between outgoing queues at line rate speeds, but arbitrating between terminated sessions at very high speeds. The stickiness of sessions across a pipeline of stages, including a general purpose OS, can be a scheduler circuit optimizing any or all such stages of such a pipeline.

Alternatively, a scheduling scheme can be used as shown in U.S. Pat. No. 7,760,715 issued to Dalal on Jul. 20, 2010, incorporated herein by reference. This scheme can be useful when it is desirable to rate limit the flows for preventing the downstream congestion of another resource specific to the over-selected flow, or for enforcing service contracts for particular flows. Embodiments can include arbitration scheme that allows for service contracts of downstream resources, such as general purpose OS that can be enforced seamlessly.

Referring still to FIG. 3, a hardware scheduler according to embodiments herein, or equivalents, can provide for the classification of incoming packet data into session flows based on session metadata. It can further provide for traffic management of these flows before they are arbitrated and queued as distinct processing entities on the offload processors.

In some embodiments, offload processors (e.g., 308i) can be general purpose processing units capable of handling packets of different application or transport sessions. Such offload processors can be low power processors capable of executing general purpose instructions. The offload processors could be any suitable processor, including but not limited to: ARM, ARC, Tensilica, MIPS, StrongARM or any other processor that serves the functions described herein. Such offload processors have a general purpose OS running on them, wherein the general purpose OS is optimized to reduce the penalty associated with context switching between different threads or group of threads.

In contrast, context switches on host processors can be computationally intensive processes that require the register save area, process context in the cache and TLB entries to be restored if they are invalidated or overwritten. Instruction Cache misses in host processing systems can lead to pipeline stalls and data cache misses lead to operation stall and such cache misses reduce processor efficiency and increase processor overhead.

Also in contrast, an OS 308j running on the offload processors 308i in association with a scheduler circuit 308b/n, can operate together to reduce the context switch overhead incurred between different processing entities running on it. Embodiments can include a cooperative mechanism between a scheduler circuit and the OS on the offload processor 308i, wherein the OS sets up session context to be physically contiguous (physically colored allocator for session heap and stack) in the cache; then communicates the session color, size, and starting physical address to the scheduler circuit upon session initialization. During an actual context switch, a scheduler circuit can identify the session context in the cache by using these parameters and initiate a bulk transfer of these contents to an external low latency memory (e.g., 308g). In addition, the scheduler circuit can manage the prefetch of the old session if its context was saved to a local memory 308g. In particular embodiments, a local memory 308g can be low latency memory, such as a reduced latency dynamic random access memory (RLDRAM), as but one very particular embodiment. Thus, in embodiments, session context can be identified distinctly in the cache.

In some embodiments, context size can be limited to ensure fast switching speeds. In addition or alternatively, embodiments can include a bulk transfer mechanism to transfer out session context to a local memory 308g. The cache contents stored therein can then be retrieved and prefetched during context switch back to a previous session. Different context session data can be tagged and/or identified within the local memory 308g for fast retrieval. As noted above, context stored by one offload processor may be recalled by a different offload processor.

In the very particular embodiment of FIG. 3, multiple offload processing cores can be integrated into a computation FPGA 308. Multiple computational FPGAs can be arbitrated by arbitrator circuits in another FPGA 310. The combination of computational FPGAs (e.g., 308) and arbiter FPGAs (e.g., 310) are referred to as “XIMM” modules or “Xockets DIMM modules” (e.g., computation unit 300). In particular applications, these XIMM modules can provide integrated traffic and thread management circuits that broker execution of multiple sessions on the offload processors.

FIG. 3 also shows an offload processor tunnel connection 308k, as well as a memory interface 308m and access unit 308l (which can be an accelerator coherency port (ACP)). Memory interface 308m can access buffer memory 308a. According to embodiments, system 301 can include use an access (or “snooping” unit) 308l, to access the cache contents of an offload processor 308i. In particular embodiments, the cache accessed can be an L2 cache. An access unit 308l can provide a port or other access capability that can load data from an external, non-cached memory 308g into an offload processor cache, as well as transfer the cache contents of an offload processor 308i to a non-cache memory 308g. As part of a computational element 300, there can be several memory devices (e.g., RAMs) that form memory 308g. Thus, memory 308g can be used to store the cache contents of sessions. Memory 308g can include one or more low latency memories, and can be conceptualized as supplementing and/or augmenting the available L2 cache, and extending the coherency domain of sessions. The addition memory 308g and access unit 308l can reduce the adverse effects of cache misses for switched in sessions, in that a session's context can be fetched and pre-fetched into an offload processor cache so that the when the thread resumes, most of its previous working set is already present in the cache.

According to one particular embodiment, in a session switch-out, an offload processor 308i cache contents can be transferred to memory 308g via tunnel 308k. However, in some embodiments, a thread's register set may be saved to memory as part of switch-out, thus these register contents can be resident in the cache. Therefore, as part of switch-in, when a session's contents are prefetched and transferred into the cache of an offload processor 308i, the register contents can be loaded by the kernel upon resuming the thread, and these loads should be from the cache and not from memory 308g. Thus, with the careful management of a session's cache contents, the cost of context switching due to register set save and restore and cache misses on switch-in can be greatly reduced, and even eliminated in some optimal cases, thereby eliminating two sources of context switch overhead and reducing the latency for the switched-in session to resume useful processing.

According to some embodiments, an access (or snooping) unit (e.g., 308l) can have the indices of all the lines in the cache where the relevant session context resides. If the session is scattered across locations in a physically indexed cache, it can become very cumbersome to access all of the session contents as multiple address translations would be required to access multiple pages of the same session. Accordingly, embodiments can include a page coloring scheme, in which the session contents are established in contiguous locations in a physically indexed cache. A memory allocator for session data can allocate from physically contiguous pages so that there is control over physical address ranges for the sessions. In some embodiments, this is done by aligning the virtual memory page and the physical memory page to index to the same location in the cache (e.g., FIG. 1-3). In alternate embodiments, virtual and physical memory pages do not have to index the same location in a physically indexed cache, but the different pages of the session can be contiguous in physical memory, such that knowledge of the beginning index and size of the entry in the cache suffices to access all session data. Further, the set size is equal to the size of a session, so that once the index of a session entry in the cache is known; the index, the size and the set color could be used to completely transfer out the session contents from the cache to an external memory (e.g., 308g).

According to embodiments, all pages of a session can be assigned the same color in a cache of an offload processor. In a particular embodiment, all pages of a session can start at the page boundary of a defined color. The number of pages allocated to a color can be fixed based on the size of a session in the cache. An offload processor (e.g., 308i) can be used for executing a specific type of sessions and it is informed of the size of each session beforehand. Based on this, the offload processor can begin a new entry at a session boundary. It can similarly allocate pages in physical memory that index to the session boundary in the cache. The entire cache context can be saved beginning at the session boundary. In the currently described embodiment, multiple pages in the session can be contiguous in a physically indexed cache. Multiple pages of a session can have the same color (i.e., they are part of the same set) and are located contiguously. Pages of a session are accessible by using an offset from the base index of the session. The cache can be arranged and broken up into distinct sets, not as pages, but as sessions. To move from one session to another, the memory allocation scheme uses an offset to the lowest bit of the indexes used to access these sessions. For example, a physically indexed cache can be an L2 cache having a size of 512 kb. The cache can be 8-way associative, with eight ways per set possible in the L2 cache. Therefore, there are eight lines per any color in L2, or eight separate instances of each color in L2. With a session context size of 8 Kb, there will then be eight different session areas within the 512 Kb L2 cache, or eight session colors with these chosen sizes.

According to embodiments, a physical memory allocator can identify the color corresponding to a session based on the cache entry/main memory entry of the temporally previous session. In a particular embodiment, the physical memory allocator can identify a session of the previous session based on the 3 bits of the address used to assign a cache entry to the previous session. The physical memory allocator can assign the new session to a main memory location (whose color can be determined through a few comparisons to the most recently used entry) and will cause a cache entry corresponding to a session of a different color to be evicted based on a least recently used policy. In another embodiment, an offload processor can include multiple cores. In such an embodiment, cache entries can be locked out for use by each processor core. For example, if the offload processor has two cores, a given set of cache lines in a cache (i.e., L2 cache) can be divided among processors, halving the number of colors. The color of the session, index of the session and session size to an external scheduler when a new session is created can be communicated. This information can be used for queue management of incoming session flows.

Embodiments can also permit isolation of shared text and any shared data by locking these lines into a cache, apart from session data. Again, a physical memory allocator and physical coloring techniques can be used. If separate shared data is placed in a cache, it is possible to lock it into the cache, as long as transfers by an access unit (e.g., ACP) do not copy such lines. When allocating memory for session data, the memory allocator can be aware of physical color, as session data that resides in the cache is mapped.

Having described various embodiments suitable for cache and context switching management operations, an example illustrating particular aspects will now be described.

FIG. 4 shows a method 400 of reduced overhead context switching for a system according to an embodiment. At initialization, a determination can be made if session coloring is required (402). Such a determination can be made by an OS. If session coloring is not required (No from 402), page coloring may or may not be present depending upon default choices of an OS (424).

If session coloring is required (Yes from 402), an OS can initializes a memory allocator (404). A memory allocator can employ cache optimization techniques that can allocate each session entry to a “session” boundary. The memory allocator can determine the starting address of each session, the number of sessions allowable in the cache, and the number of locations wherein a session can be found for a given color. Such operations can include determining the number of sets available based cache size, number of colors and size of a session (step 406).

When a packet for a session arrives, a determination can be made of whether the packet is for a current or different session (408). Such an action can be performed by the OS. If a packet is for a different session (Yes from 408), a determination can be made of whether the packet is form an earlier session (410). If the packet is not from an earlier session (i.e., it is a new session), a determination can be made to see if there is enough memory for the new session (418). If there is enough space (Yes from 418), a switch can be made to the new session (422). Such an action can include allocating a new session at a session boundary, and saving the context of the process that is currently executing to a context memory (which can be an external low latency memory).

If no cache memory is available for a new session (No from 418) and/or the packet is for an earlier session (Yes from 410) an examination can be made to determine if the packet of the old/new session is of a same color (412). If it is of a different color (No from 412), a switch can be made to that session (step 414). Such an action can include retrieving (for an earlier session) or creating (for a new session) cache entries for the task. In addition, such an action can include, if needed, the flushing of cache entries according to an LRU scheme.

If a packet of the old/new session is of the same color (Yes from 412), a determination can be made as to whether the color pressure can be exceeded (416). If the color pressure can be exceeded, or a session of some other color is not available (Yes, or . . . from 416), a switch can be made to the new session (420). Such an action can include creating cache entries and remembering the new session color. If the color pressure cannot be exceeded, but sessions of some other color are available (No, but . . . from 416), a method can proceed to 414.

It should be appreciated that in the foregoing description of exemplary embodiments of the invention, various features of the invention are sometimes grouped together in a single embodiment, figure, or description thereof for the purpose of streamlining the disclosure aiding in the understanding of one or more of the various inventive aspects. This method of disclosure, however, is not to be interpreted as reflecting an intention that the claimed invention requires more features than are expressly recited in each claim. Rather, as the following claims reflect, inventive aspects lie in less than all features of a single foregoing disclosed embodiment. Thus, the claims following the detailed description are hereby expressly incorporated into this detailed description, with each claim standing on its own as a separate embodiment of this invention.

It is also understood that the embodiments of the invention may be practiced in the absence of an element and/or step not specifically disclosed. That is, an inventive feature of the invention may be elimination of an element.

Accordingly, while the various aspects of the particular embodiments set forth herein have been described in detail, the present invention could be subject to various changes, substitutions, and alterations without departing from the spirit and scope of the invention.

Claims

1. A method for context switching of multiple offload processors, comprising the steps of:

receiving network packets for processing through a memory bus connected socket,
organizing the network packets into multiple sessions for processing,
suspending processing of at least one session by reading a cache state of at least one of the offload processor into a context memory by operation of a scheduling circuit, with virtual memory locations and physical cache locations being aligned, and
subsequently directing transfer of the cache state to at least one of the offload processors for processing by operation of the scheduling circuit.

2. The method of claim 1 wherein the bulk read is through an accelerator coherency port.

3. The method of claim 1 wherein the associated cache state includes at least one of: a state of offload processor registers, instructions for execution by the offload processor, a stack pointer, program counter, prefetched instructions for execution by the offload processor, prefetched data for use by the offload processor, and data written into the cache of the offload processor.

4. The method of claim 1, further including:

the cache state includes session context; and
setting the session context to be physically contiguous in the cache of the offload processor.

5. The method of claim 4, wherein the setting of the session context includes cooperation of an operating system (OS) running on the offload processor and the scheduling circuit.

6. The method of claim 1, further including upon initialization of a processing session, communicating session data to the scheduling circuit.

7. The method of claim 6, wherein the session data includes any selected from: a session color, a session size and starting physical cache address for the processing session.

8. The method of claim 1, further including

determining starting address for each of a plurality of processing sessions, the number of sessions allowable in a cache of an offload processor, and the number of locations wherein a session can be found for a given session color.

9. The method of claim 1 further including

transferring the cache state of one of the offload processors to the cache of another of the offload processors.

10. The method of claim 1, further including prioritizing a processing of network packets in a first queue received over the memory bus by stopping a first session associated with one of the offload processors, storing the cache state of the offload processor, and initiating processing of network packets held in a second queue.

11. The method of claim 1, wherein the suspending processing of at least one session includes operating in a preemption mode to control a session execution.

12. The method of claim 1, wherein receiving network packets includes receiving network packets through a dual in line memory module (DIMM) compatible socket.

13. The method of claim 1, wherein reading the cache state into the context memory includes reading the cache state data over the memory bus.

Patent History

Publication number: 20140201761
Type: Application
Filed: Jun 26, 2013
Publication Date: Jul 17, 2014
Inventors: Parin Bhadrik Dalal (Milpitas, CA), Stephen Paul Belair (Santa Cruz, CA)
Application Number: 13/928,339

Classifications

Current U.S. Class: Context Switching (718/108)
International Classification: G06F 9/46 (20060101);