Technologies for processing network packets by a network interface controller (NIC) of a computing device include a network interface, a packet processor, and a controller device of the NIC, each communicatively coupled to a memory fabric of the NIC. The packet processor is configured to receive an event message from the memory fabric and transmit a message to the controller device, wherein the message indicates the network packet has been received and includes the memory fabric location pointer. The controller device is configured to fetch at least a portion of the received network packet from the memory fabric, write an inbound descriptor usable by one or more on-die cores of the NIC to perform an operation on the fetched portion, and restructure the network packet as a function of an outbound descriptor written by the on-die cores subsequent to performing the operation. Other embodiments are described herein.

A computer network organized in a logical grid having rows and columns can include network nodes coupled according to harmonics. Each network node can be coupled to network nodes of the same row using a set of horizontal strands according to a set of horizontal harmonics. Each of the horizontal harmonics specifies a node distance along the row between adjacent connection points on the corresponding horizontal strand. Each network node can also be coupled to network nodes of the same column using a set of vertical strands according to a set of vertical harmonics. Each of the vertical harmonics specifies a node distance along the column between adjacent connection points on the corresponding vertical strand.

The subject technology relates to the management of a shared buffer memory in a network switch. Systems, methods, and machine readable media are provided for receiving a data packet at a first network queue from among a plurality of network queues, determining if a fill level of a queue in a shared buffer of the network switch exceeds a dynamic queue threshold, and in an event that the fill level of the shared buffer exceeds the dynamic queue threshold, determining if a fill level of the first network queue is less than a static queue minimum threshold.

Technologies for offloading acceleration task scheduling operations to accelerator sleds include a compute device to receive a request from a compute sled to accelerate the execution of a job, which includes a set of tasks. The compute device is also to analyze the request to generate metadata indicative of the tasks within the job, a type of acceleration associated with each task, and a data dependency between the tasks.

Technologies for offloading acceleration task scheduling operations to accelerator sleds include a compute device to receive a request from a compute sled to accelerate the execution of a job, which includes a set of tasks. The compute device is also to analyze the request to generate metadata indicative of the tasks within the job, a type of acceleration associated with each task, and a data dependency between the tasks.

Technologies for dividing work across one or more accelerator devices include a compute device. The compute device is to determine a configuration of each of multiple accelerator devices of the compute device, receive a job to be accelerated from a requester device remote from the compute device, and divide the job into multiple tasks for a parallelization of the multiple tasks among the one or more accelerator devices, as a function of a job analysis of the job and the configuration of each accelerator device. The compute engine is further to schedule the tasks to the one or more accelerator devices based on the job analysis and execute the tasks on the one or more accelerator devices for the parallelization of the multiple tasks to obtain an output of the job.

Technologies for dividing work across one or more accelerator devices include a compute device. The compute device is to determine a configuration of each of multiple accelerator devices of the compute device, receive a job to be accelerated from a requester device remote from the compute device, and divide the job into multiple tasks for a parallelization of the multiple tasks among the one or more accelerator devices, as a function of a job analysis of the job and the configuration of each accelerator device. The compute engine is further to schedule the tasks to the one or more accelerator devices based on the job analysis and execute the tasks on the one or more accelerator devices for the parallelization of the multiple tasks to obtain an output of the job.

An endpoint group (EPG) can be stretched between the sites so that endpoints at different sites can be assigned to the same stretched EPG. Because the sites can use different bridge domains when establishing the stretched EPGs, the first time a site transmits a packet to an endpoint in a different site, the site learns or discovers a path to the destination endpoint. The site can use BGP to identify the site with the host and use a multicast tunnel to reach the site. A unicast tunnel can be used to transmit future packets to the destination endpoint. Additionally, a stretched EPG can be segmented to form a micro-stretched EPG. Filtering criteria can be used to identify a subset of the endpoints in the stretched EPG that are then assigned to the micro-stretched EPG, which can have different policies than the stretched EPG.

An endpoint group (EPG) can be stretched between the sites so that endpoints at different sites can be assigned to the same stretched EPG. Because the sites can use different bridge domains when establishing the stretched EPGs, the first time a site transmits a packet to an endpoint in a different site, the site learns or discovers a path to the destination endpoint. The site can use BGP to identify the site with the host and use a multicast tunnel to reach the site. A unicast tunnel can be used to transmit future packets to the destination endpoint. Additionally, a stretched EPG can be segmented to form a micro-stretched EPG. Filtering criteria can be used to identify a subset of the endpoints in the stretched EPG that are then assigned to the micro-stretched EPG, which can have different policies than the stretched EPG.

One aspect of the instant application can provide a networking device. The networking device can include a set of Compute Express Link (CXL) ports, a set of non-CXL ports, a set of bridge circuits associated with the set of CXL ports, and an interconnect. A respective bridge circuit can include a packet-conversion subcircuit to convert a CXL packet received at a corresponding CXL port to a customized packet and a protocol-conversion subcircuit to convert a CXL protocol to a customized protocol implemented by the networking device. The interconnect can switch customized packets among the CXL and non-CXL ports based on the customized protocol.