摘要:
Technologies for providing accelerated functions as a service in a disaggregated architecture include a compute device that is to receive a request for an accelerated task. The task is associated with a kernel usable by an accelerator sled communicatively coupled to the compute device to execute the task. The compute device is further to determine, in response to the request and with a database indicative of kernels and associated accelerator sleds, an accelerator sled that includes an accelerator device configured with the kernel associated with the request. Additionally, the compute device is to assign the task to the determined accelerator sled for execution. Other embodiments are also described and claimed.
摘要:
A method for inverse multiplexing of managed and unmanaged traffic flows over a multi-star switch network includes a source node classifier (410). The source node classifier, using a traffic-engineering algorithm, classifies incoming traffic based on flow parameters, embeds the flow parameters in a routing table in a node for a flow, places packets from classified flows into Switch-Specific Managed-Traffic Queues (SSMT 430) and a source node unmanaged traffic queue (440). A source node switch input scheduler process for a switch (450…452) selects all managed packets from the SSMT destined for the switch, then selects a single unmanaged packet from the source node unmanaged traffic queue. The source node transmits the packets as classified flows through a switch fabric to the destination node. At the destination mode packets transmitted through the switch fabric are sorted by a Switch output Process and sent to intended output queues (420…422).
摘要:
A method for inverse multiplexing of managed and unmanaged traffic flows over a multi-star switch network includes a source node classifier (410). The source node classifier, using a traffic-engineering algorithm, classifies incoming traffic based on flow parameters, embeds the flow parameters in a routing table in a node for a flow, places packets from classified flows into Switch-Specific Managed-Traffic Queues (SSMT 430) and a source node unmanaged traffic queue (440). A source node switch input scheduler process for a switch (450…452) selects all managed packets from the SSMT destined for the switch, then selects a single unmanaged packet from the source node unmanaged traffic queue. The source node transmits the packets as classified flows through a switch fabric to the destination node. At the destination mode packets transmitted through the switch fabric are sorted by a Switch output Process and sent to intended output queues (420…422).
摘要:
A method and system for a network node for attachment to switch fabrics is described. The system includes an access unit to provide access to communications from an external network, a classification element to label received packets with information identifying an associated flow and queue, a mapping element to place the packets into one of a plurality of queues based on the label identifiers, a scheduler to schedule packets in the queues for transmission, and an encapsulation element to encapsulate the scheduled packets into uniform size frames. The uniform size frames may then be transmitted to a next destination through a switch fabric.
摘要:
A method and system for a network node for attachment to switch fabrics is described. The system includes an access unit to provide access to communications from an external network, a classification element to label received packets with information identifying an associated flow and queue, a mapping element to place the packets into one of a plurality of queues based on the label identifiers, a scheduler to schedule packets in the queues for transmission, and an encapsulation element to encapsulate the scheduled packets into uniform size frames. The uniform size frames may then be transmitted to a next destination through a switch fabric.
摘要:
Network switch circuitry for use with a physical network infrastructure is provided that supports multiple different link-layer protocols and that provides telemetry information related to the network switch circuitry. A method and a machine-readable storage medium are also provided.
摘要:
Technologies for providing efficient reprovisioning in an accelerator device include an accelerator sled. The accelerator sled includes a memory and an accelerator device coupled to the memory. The accelerator device is to configure itself with a first bit stream to establish a first kernel, execute the first kernel to produce output data, write the output data to the memory, configure itself with a second bit stream to establish a second kernel, and execute the second kernel with the output data in the memory used as input data to the second kernel. Other embodiments are also described and claimed.
摘要:
A method and system for open-loop congestion control in a system fabric is described. The method includes determining which traffic class each received network packet belongs, determining a path to be taken by each packet through a switch fabric, classifying each packet into one of a plurality of flow bundles based on the packet’s destination and path through the switch fabric, mapping each packet into one of a plurality of queues to await transmission based on the flow bundle to which the packet has been classified, and scheduling the packets in the queues for transmission to a next destination through the switch fabric.
摘要:
Multiple voice channels are aggregated into a packet having a segmented data structure and sent over a packet network. The segmented data structure includes segment IDs, corresponding data segments and a packet header arranged so that all data is aligned on 8-byte boundaries for efficient processing by 64-bit processors. The data segment represents one or more milliseconds of digitized voice data, and the segment ID explicitly identifies the voice channel associated with the digitized voice data without reference to any other data in the data structure.