A network element (101), for example a router, comprises a processing system (103) for running a protocol in accordance of which the network element is communicating with another network element. The protocol comprises monitoring whether the protocol receives, within a protocol message -specific time-window, a protocol message transmitted by the other network element. The time-window may represent for example a timeout period. The processing system measures waiting times of messages received at the network element, where each measured waiting time is at least a part of time a corresponding received message waits for a service provided by the processing system. The processing system determines the length of the time-window based on the measured waiting times. Therefore, the waiting time of the received protocol message can be compensated for and thereby wrong determinations about the reachability of the other network element can be avoided or at least reduced.

Each of a plurality of input units (110) outputs an input notification (104) at the input timing at which input data (101) is input and outputs input data at the synchronization timing which is the timing of the later one of the input timing and a timing at which an input notification is output from another input unit. Each of a plurality of computing units (120) starts a computation when input data output from an associated input unit is input.

A switch fabric modification system includes a switch fabric having first switch devices and a primary switch device. The primary switch device detects a second switch device has been connected to the switch fabric and, in response and without subsequently receiving an instruction from a user, verifies a second switch device/switch fabric connectivity of the second switch device to the switch fabric that provides a switch device role for the second switch device by determining that the second switch device/switch fabric connectivity enables data communications that are available to the first switch device(s) that have the switch device role, verifies an operating configuration of the second switch device by determining that it matches an operating configuration of the primary switch device and, response to verifying the second switch device/switch fabric connectivity and the operating configuration of the second switch device, adds the second switch device to the switch fabric.

Discussed herein is a mechanism of building/constructing a network fabric for a cluster of GPUs. A plurality of sets of GPUs are created, wherein each set of GPUs is created by selecting one GPU from each host machine in the plurality of host machines. Each set of GPUs is coupled to a different group of switches in a plurality of groups of switches. The coupling included: (i) coupling each GPU in the set of GPUs to a unique ingress port of a first switch included in a corresponding group of switches that is associated with the set of GPUs, and (ii) mapping virtually, each ingress port of the first switch to a unique egress port of a plurality of egress ports of the first switch. A packet originating at a source GPU and destined for a destination GPU is communicated via the network fabric.