Abstract: Approaches, techniques, and mechanisms are disclosed for maintaining efficient representations of prefix tables for utilization by network switches and other devices. In an embodiment, the performance of a network device is greatly enhanced using a working representation of a prefix table that includes multiple stages of prefix entries. Higher-stage prefixes are stored in slotted pools. Mapping logic, such as a hash function, determines the slots in which a given higher-stage prefix may be stored. When trying to find a longest-matching higher-stage prefix for an input key, only the slots that map to that input key need be read. Higher-stage prefixes may further point to arrays of lower-stage prefixes. Hence, once a longest-matching higher-stage prefix is found for an input key, the longest prefix match in the table may be found simply by comparing the input key to lower-stage prefixes in the array that the longest-matching higher-stage prefix points to.

Abstract: Automatic load-balancing techniques in a network device are used to select, from a multipath group, a path to assign to a flow based on observed state attributes such as path state(s), device state(s), port state(s), or queue state(s) of the paths. A mapping of the path previously assigned to a flow or group of flows (e.g., on account of having then been optimal in view of the observed state attributes) is maintained, for example, in a table. So long as the flow(s) are active and the path is still valid, the mapped path is selected for subsequent data units belonging to the flow(s), which may, among other effects, avoid or reduce packet re-ordering. However, if the flow(s) go idle, or if the mapped path fails, a new optimal path may be assigned to the flow(s) from the multipath group.

Abstract: Distributed machine learning systems and other distributed computing systems are improved by embedding compute logic at the network switch level to perform collective actions, such as reduction operations, on gradients or other data processed by the nodes of the system. The switch is configured to recognize data units that carry data associated with a collective action that needs to be performed by the distributed system, referred to herein as “compute data,” and process that data using a compute subsystem within the switch. The compute subsystem includes a compute engine that is configured to perform various operations on the compute data, such as “reduction” operations, and forward the results back to the compute nodes. The reduction operations may include, for instance, summation, averaging, bitwise operations, and so forth. In this manner, the network switch may take over some or all of the processing of the distributed system during the collective phase.

Abstract: In response to certain events in a network device, visibility packets may be generated. A visibility packet may be or comprise at least a portion of a packet that is in some way associated with the event, such as a packet that was dropped as a result of an event, or that was in a queue at the time of an event related to that queue. A visibility packet may be tagged or otherwise indicated as a visibility packet. The network device may include one or more visibility samplers through which visibility packets are routed on their way to a visibility queue, visibility subsystem, and/or out of the network device. The samplers allow only a limited sample of the visibility packets that they receive to pass through the sampler, essentially acting as a filter to reduce the amount of visibility packets that will be processed.

Abstract: Distributed machine learning systems and other distributed computing systems are improved by compute logic embedded in extension modules coupled directly to network switches. The compute logic performs collective actions, such as reduction operations, on gradients or other compute data processed by the nodes of the system. The reduction operations may include, for instance, summation, averaging, bitwise operations, and so forth. In this manner, the extension modules may take over some or all of the processing of the distributed system during the collective phase. An inline version of the module sits between a switch and the network. Data units carrying compute data are intercepted and processed using the compute logic, while other data units pass through the module transparently to or from the switch. Multiple modules may be connected to the switch, each coupled to a different group of nodes, and sharing intermediate results. A sidecar version is also described.

Abstract: Approaches, techniques, and mechanisms are disclosed for assigning paths to network packets. The path assignment techniques utilize path state information and/or other criteria to determine whether to route a packet along a primary candidate path selected for the packet, or one or more alternative candidate paths selected for the packet. According to an embodiment, network traffic is at least partially balanced by redistributing only a portion of the traffic that would have been assigned to a given primary path. Move-eligibility criteria are applied to traffic to determine whether a given packet is eligible for reassignment from a primary path to an alternative path. The move-eligibility criteria determine which portion of the network traffic to move and which portion to allow to proceed as normal. In an embodiment, the criteria and functions used to determine whether a packet is redistributable are adjusted over time based on path state information.

Abstract: Nodes within a network are configured to adapt to changing path states, due to congestion, node failures, and/or other factors. A node may selectively convey path information and/or other state information to another node by annotating the information into packets it receives from the other node. A node may selectively reflect these annotated packets back to the other node, or other nodes that subsequently receive these annotated packets may reflect them. A weighted cost multipathing selection technique is improved by dynamically adjusting weights of paths in response to feedback indicating the current state of the network topology, such as collected through these reflected packets. In an embodiment, certain packets that would have been dropped may instead be transformed into “special visibility” packets that may be stored and/or sent for analysis. In an embodiment, insight into the performance of a network device is enhanced through the use of programmable visibility engines.

Abstract: Approaches, techniques, and mechanisms are disclosed for reutilizing discarded link data in a buffer space for buffering data units in a network device. Rather than wasting resources on garbage collection of such link data when a data unit is dropped, the link data is used as a free list that indicates buffer entries in which new data may be stored. In an embodiment, operations of the buffer may further be enhanced by re-using the discarded link data as link data for a new data unit. The link data for a formerly buffered data unit may be assigned exclusively to a new data unit, which uses the discarded link data to determine where to store its constituent data. As a consequence, the discarded link data actually serves as valid link data for the new data unit, and new link data need not be generated for the new data unit.

Abstract: Approaches, techniques, and mechanisms facilitate actionable reporting of network state information and real-time, autonomous network engineering directly in-network at a switch or other network device. A data collector within the network device collects state information and/or data unit information from various device components, such as traffic managers and packet processors. The data collector, which may optionally generate additional state information by performing various calculations on the information it receives, is configured to then provide at least some of the state information to an analyzer device connected to an analyzer interface. The analyzer device, which may be a separate device, performs various analyses on the state information, depending on how it is configured. The analyzer device outputs reports that identify statuses, errors, misconfigurations, and/or suggested actions to take to improve operation of the network device.

Abstract: Approaches, techniques, and mechanisms are disclosed for efficiently buffering data units within a network device. A traffic manager or other network device component receives Transport Data Units (“TDUs”), which are sub-portions of Protocol Data Units (“PDUs”). Rather than buffer an entire TDU together, the component divides the TDU into multiple Storage Data Units (“SDUs”) that can fit in SDU buffer entries within physical memory banks. A TDU-to-SDU Mapping (“TSM”) memory stores TSM lists that indicate which SDU entries store SDUs for a given TDU. Physical memory banks in which the SDUs are stored may be grouped together into logical SDU banks that are accessed together as if a single bank. The TSM memory may include a number of distinct TSM banks, with each logical SDU bank having a corresponding TSM bank. Techniques for maintaining inter-packet and intra-packet linking data compatible with such buffers are also disclosed.

Abstract: Distributed machine learning systems and other distributed computing systems are improved by embedding compute logic at the network switch level to perform collective actions, such as reduction operations, on gradients or other data processed by the nodes of the system. The switch is configured to recognize data units that carry data associated with a collective action that needs to be performed by the distributed system, referred to herein as “compute data,” and process that data using a compute subsystem within the switch. The compute subsystem includes a compute engine that is configured to perform various operations on the compute data, such as “reduction” operations, and forward the results back to the compute nodes. The reduction operations may include, for instance, summation, averaging, bitwise operations, and so forth. In this manner, the network switch may take over some or all of the processing of the distributed system during the collective phase.

Abstract: Distributed machine learning systems and other distributed computing systems are improved by embedding compute logic at the network switch level to perform collective actions, such as reduction operations, on gradients or other data processed by the nodes of the system. The switch is configured to recognize data units that carry data associated with a collective action that needs to be performed by the distributed system, referred to herein as “compute data,” and process that data using a compute subsystem within the switch. The compute subsystem includes a compute engine that is configured to perform various operations on the compute data, such as “reduction” operations, and forward the results back to the compute nodes. The reduction operations may include, for instance, summation, averaging, bitwise operations, and so forth. In this manner, the network switch may take over some or all of the processing of the distributed system during the collective phase.

Abstract: An improved buffer for networking devices and other computing devices comprises multiple memory instances, each having a distinct set of entries. Transport data units (“TDUs”) are divided into storage data units (“SDUs”), and each SDU is stored within a separate entry of a separate memory instance in a logical bank. A grid of the memory instances is organized into overlapping horizontal logical banks and vertical logical banks. A memory instance may be shared between horizontal and vertical logical banks. When overlapping logical banks are accessed concurrently, the memory instance that they share may be inaccessible to one of the logical banks. Accordingly, when writing a TDU, a parity SDU may be generated for the TDU and also stored within its logical bank. The TDU's content within the shared memory instance may then be reconstructed from the parity SDU without having to read the shared memory instance.

Abstract: When a measure of buffer space queued for garbage collection in a network device grows beyond a certain threshold, one or more actions are taken to decreasing an enqueue rate of certain classes of traffic, such as of multicast traffic, whose reception may have caused and/or be likely to exacerbate garbage-collection-related performance issues. When the amount of buffer space queued for garbage collection shrinks to an acceptable level, these one or more actions may be reversed. In an embodiment, to more optimally handle multi-destination traffic, queue admission control logic for high-priority multi-destination data units, such as mirrored traffic, may be performed for each destination of the data units prior to linking the data units to a replication queue. If a high-priority multi-destination data unit is admitted to any queue, the high-priority multi-destination data unit can no longer be dropped, and is linked to a replication queue for replication.

Abstract: To more efficiently utilize buffer resources, schedulers within a traffic manager may generate and queue read instructions for reading buffered portions of data units that are ready to be sent to the egress blocks. The traffic manager may be configured to select a read instruction for a given buffer bank from the read instruction queues based on a scoring mechanism or other selection logic. To avoid sending too much data to an egress block during a given time slot, once a data unit portion has been read from the buffer, it may be temporarily stored in a shallow read data cache. Alternatively, a single, non-bank specific controller may determine all of the read instructions and write operations that should be executed in a given time slot. The read instruction queue architecture may be duplicated for link memories and other memories in addition to the buffer memory.

Abstract: A logical bank may comprise multiple physical banks across which logical blocks, such as buffer entries, are striped. Data structures are stored in the logical blocks, with each logical block storing no more than one data structure. When writing a data structure, if the data structure is less than half the logical block size, one or more duplicate copies of the data structure may be stored in the otherwise unused physical blocks of the logical block. Before executing a first read instruction to read a first data structure from a first logical block, if the first data structure can be read without accessing one or more of the physical banks, a second read instruction for a second data structure may be analyzed to determine if there is a copy of a second data structure within the one or more unneeded physical banks. If so, the two read instructions are consolidated.

Abstract: Certain hash-based operations in network devices and other devices, such as mapping and/or lookup operations, are improved by manipulating a hash key prior to executing a hash function on the hash key and/or by manipulating outputs of a hash function. A device may be configured to manipulate hash keys and/or outputs using manipulation logic based on one or more predefined manipulation values. A similar hash-based operation may be performed by multiple devices within a network of computing devices. Different devices may utilize different predefined manipulation values for their respective implementations of the manipulation logic. For instance, each device may assign itself a random mask value for key transformation logic as part of an initialization process when the device powers up and/or each time the device reboots. In an embodiment, described techniques may increase the entropy of hashing function outputs in certain contexts, thereby increasing the effectiveness of certain hashing functions.

Abstract: Methods, systems, and apparatus, including a managed device comprising memory storage, one or more control registers, and circuitry to perform operations of receiving, from a control system, one or more posted write operations directed to the one or more control registers; based on the one or more posted write operations, storing in the one or more control registers, data specifying at least a system address of a memory of the control system, where the system address corresponds to a starting address of a predetermined section of the memory; and transferring managed device data from the memory storage to the predetermined section of the memory of the control system by issuing, to the control system and based on the system address of the memory, one or more posted write operations to write the managed device data to the predetermined section of the memory.

Abstract: A memory system for a network device is described. The memory system includes a main memory configured to store one or more data elements. Further, the memory system includes a link memory that is configured to maintain one or more pointers to interconnect the one or more data elements stored in the main memory. The memory system also includes a free-entry manager that is configured to generate an available bank set including one or more locations in the link memory. In addition, the memory system includes a context manager that is configured to maintain metadata for a list of the one or more data elements.

Abstract: A traffic manager is shared amongst two or more egress blocks of a network device, thereby allowing traffic management resources to be shared between the egress blocks. Among other aspects, this may reduce power demands and allow a larger amount of buffer memory to be available to a given egress block that may be experiencing high traffic loads. Optionally, the shared traffic manager may be leveraged to reduce the resources required to handle data units on ingress. Rather than buffer the entire unit in the ingress buffers, an arbiter may be configured to buffer only the control portion of the data unit. The payload of the data unit, by contrast, is forwarded directly to the shared traffic manager, where it is placed in the egress buffers. Because the payload is not being buffered in the ingress buffers, the ingress buffer memory may be greatly reduced.