Abstract: Allocating bandwidth in a network (100) may include reducing initial bandwidth allocations for each tenant on a link (113) such that a sum of the bandwidth allocations is equal to or less than a capacity of the link and distributing a residual of the initial bandwidth allocations to at least one under-allocated link (114).

Abstract: Connectors of a first removable modular optical connection assembly, having a first predefined arrangement of optical signal conduits, are connected to respective connectors on a support structure that are optically connected to corresponding devices. The first modular optical connection assembly is replaceable with a second modular optical connection assembly having a second, different predefined arrangement of optical signal conduits, to change a topology of a network.

Abstract: A network switch for reducing TCP timeouts due to incast collapse is disclosed. The network switch has a plurality of buffers for storing network packets for at least one sender server. An incast buffer management component in the network switch allocates a buffer in the plurality of hullers to store an incast notification packet, the incast notification packet indicating one or mote network packets that have been dropped by the network switch. The network switch also includes an incast notification component for forwarding the incast notification packet to a destination server with an incast notification packet specific header.

Abstract: A scalable, multi-tenant network architecture for a virtualized datacenter is provided. The network architecture includes a network having a plurality of servers connected to a plurality of switches. The plurality of servers hosts a plurality of virtual interfaces for a plurality of tenants. A configuration repository is connected to the network and each server in the plurality of servers has a network agent hosted therein. The network agent encapsulates packets for transmission across the network from a source virtual interface to a destination virtual interface in the plurality of virtual interfaces for a tenant in the plurality of tenants. The packets are encapsulated with information identifying and locating the destination virtual interface, and the information is interpreted by switches connected to the source virtual interface and the destination virtual interface.

Abstract: Selected ones of multiple groups of network entities are assigned to corresponding routing structures, where each of the routing structures defines a set of paths through a network among network entities of a corresponding one of the groups. The assigning considers available multicast resources in communication nodes in the network, and a criterion relating to reducing overall cost in the assignment of the selected groups to the corresponding routing structures.

Abstract: A datacenter network, method, and non-transitory computer readable medium for reducing cabling costs in the datacenter network are provided. The datacenter network is represented by a network topology that interconnects a plurality of network elements and a physical topology that is organized into a plurality of physical elements and physical units. A network design module assigns network elements to the plurality of physical elements and physical units based on a hierarchical partitioning of the physical topology and a matching hierarchical partitioning of the network topology that reduces costs of cables used to interconnect the network elements in the physical topology.

Abstract: Local rules for managing flows devolved from a central controller are received at a switch. The central controller determines a global set of rules for managing flows. The switch receives a packet from a flow from a network and determines whether a metric for the flow satisfies a dynamic condition to trigger a metric report to the central controller. In response to a determination that the metric for the flow at the switch satisfies the dynamic condition to trigger a metric report to the central controller, the switch sends a metric report to the central controller, and the switch then receives an instruction to manage the flow from the central controller. In response to a determination that the metric for the flow at the switch does not satisfy the dynamic condition to trigger the metric report to the central controller, the switch manages the flow using the local rules for managing flows.

Abstract: A method of generating a plurality of potential network topologies is provided herein. The method includes receiving parameters that specify a number of servers, a number of switches, and a number of ports in the switches. The parameters are for configuring a network topology. The method also includes generating one or more potential network topologies comprising the set of potential network topologies, for each of a number of dimensions. The number of dimensions is based on the number of switches. The method further includes determining that the set of potential network topologies is structurally feasible. Additionally, the method includes determining an optimal link aggregation (LAG) factor in each dimension of each of the set of potential network topologies.

Abstract: A method of generating a plurality of potential generalized fat tree network (GFT) topologies is provided herein. The method includes receiving parameters that specify a number of servers, a number of switches, and a number of ports in the switches. The parameters are for configuring an GFT topology. The method also includes generating a plurality of potential GFT topologies based on heuristics that favor higher bandwidth and lower latency. Generating the plurality of potential GFT topologies includes generating each level of the potential GFT topologies, recursively, and from a bottom level of a potential GFT topology to a top level.

Abstract: Example methods, apparatus, and articles of manufacture to access memory are disclosed. A disclosed example method involves receiving at least one runtime characteristic associated with accesses to contents of a memory page and dynamically adjusting a memory fetch width for accessing the memory page based on the at least one runtime characteristic.

Abstract: Example methods, apparatus, and articles of manufacture to control generation of memory access requests in processor systems are disclosed. A disclosed example method involves determining at a memory controller whether a memory access queue depth for a memory reference is greater than a first threshold. When the memory access queue depth for the memory reference is greater than the first threshold, an indication is sent to control generation of memory access requests corresponding to the memory reference.

Abstract: Proactive sending of an IP-to-MAC address binding for a node is performed if a node is determined to be a high fan-in node. If the node is determined to be a high fan-in node, the node periodically sends its IP-to-MAC address binding to nodes in a network. If the node is not determined to be a high fan-in node, the node sends its IP-to-MAC address binding in response to receiving a request for the node's IP-to-MAC address binding.

Abstract: One embodiment is main memory that includes a combination of non-volatile memory (NVM) and dynamic random access memory (DRAM). An operating system migrates data between the NVM and the DRAM.

Abstract: Local rules for managing flows devolved from a central controller are received at a switch. The central controller determines a global set of rules for managing flows. The switch receives a packet from a flow from a network and determines whether a metric for the flow satisfies a dynamic condition to trigger a metric report to the central controller. In response to a determination that the metric for the flow at the switch satisfies the dynamic condition to trigger a metric report to the central controller, the switch sends a metric report to the central controller, and the switch then receives an instruction to manage the flow from the central controller. In response to a determination that the metric for the flow at the switch does not satisfy the dynamic condition to trigger the metric report to the central controller, the switch manages the flow using the local rules for managing flows.

Abstract: One embodiment is main memory that includes a combination of non-volatile memory (NVM) and dynamic random access memory (DRAM). An operating system migrates data between the NVM and the DRAM.

Abstract: A system framework for supporting system performance prediction that includes a system having a plurality of external interactors interfacing thereto, each external operator operates to service a plurality of items for input to the system, the system operates to determine arrival rates of the plurality of items arriving at the plurality of external interactors and to calculate a service time that each of the plurality of external interactors takes to service those items it receives; and a queue modeling module that operates to select a queuing model based on the received arrival times and the calculated service times and to calculate a queue length of items at each of the external interactors.

Abstract: A computer system uses a prefetch prediction model having energy usage parameters to predict the impact of prefetching specified files on the system's energy usage. A prefetch prediction engine utilizes the prefetch prediction model to evaluate the specified files with respect to prefetch criteria, including energy efficiency prefetch criteria, and generates a prefetch decision with respect to each file of the specified files. For each specified file for which the prefetch prediction engine generates an affirmative prefetch decision, an identifying entry is stored in a queue. The computer system fetches files identified by entries in the queue, although some or all of the entries in the queue at any one time may be deleted if it is determined that the identified files are no longer likely to be needed by the computer system.

Abstract: Undesirable behavior patterns of computers on a network impact network performance. A system and method are provided for limiting the impact of undesirable behavior of computers on the network. The network, through which packets of data are interchanged between the computers, includes one or more forwarding devices that are controlled or instructed by one or more packet traffic monitors. Each of the packet traffic monitors is configured for monitoring the packets; for determining if the information about the pattern of behavior from any of the computers is trustworthy; for determining, upon discovering that one or more of the patterns of behavior is undesirable, a type of the undesirable pattern behavior; and for determining a proper action for mitigating that type of undesirable behavior. The proper action is performed by mitigation means controlling the one or more forwarding devices.

Abstract: Method and apparatus for determining causal relations between a plurality of intercommunicating nodes. In various embodiments, trace data is input that describe inter-node communication. The trace data may include for each message sent between nodes a timestamp that indicates a time at which the message was sent, a source identifier that identifies a node from which the message was sent, and a destination identifier that identifies a node to which the message was sent. For each of one or more nodes, a determination may be made as to whether one or more causal relations exist between a first set of messages destined to the node and a second set of messages sourced from the node and destined to at least one other node. A causal relation may exist as a function of a probability distribution of delay values that are differences between timestamps of messages in the second set and timestamps of messages in the first set. From the nodes and causal relations a processor-readable representation is generated.

Abstract: Information returned by a server to a client includes instructions, executable by either a proxy server or the client, for converting the returned information from a first representation to a second representation. The representation conversion may be made by a proxy server, for example, to make transmission of the returned information to the client more efficient, and/or to render the returned information in a format suitable for display by the client. By having the server embed representation conversion information in the query response, the representation conversion can take into account the type and other characteristics of information being returned, as well as the computational and display characteristics of the client.