An on-chip ring network anti-starvation processing method based on fixed time slice cooperative flow control

Abstract

The invention relates to a ring network on chip anti-starvation treatment method by use of coordinating flow control in fixed time slices. An independent flow control ring network is constructed besides a message ring network to transfer flow control packages of all starvation nodes; all the starvation nodes in a same ring network direction can only send original flow control packages to their upstream nodes in fixed time slices; the original flow control packages are successively transferred to all the upstream nodes corresponding to the starvation nodes through the flow control ring network; the upstream nodes selects a proper time to forbid sending local-to-remote information of certain stride distance on the basis of the received flow control package information; a network idle time splice is cooperatively reserved so that the starvation nodes can send inserting information to relieve starvation. By use of the method, the starvation problem in a non-blocking ring network having unconditional priority of remote request is solved, and the message transfer efficiency is guaranteed to the most.

Description

technical field [0001] The invention relates to the technical field of ring networks with on-chip multi-core structure, in particular to an on-chip ring network anti-starvation processing method with fixed time slice cooperative flow control. Background technique [0002] With the advancement of semiconductor technology and the continuous improvement of System On Chip (SoC) technology, more and more processor cores, on-chip caches, memory controllers and various I / O interfaces are integrated on a single chip. The design complexity of the Network On Chip (NoC) of these modules is constantly increasing, and they are facing new challenges in many aspects such as correctness verification, back-end physical implementation, and frequency increase. [0003] The design of NoC includes several aspects such as topology structure, routing algorithm, service quality and congestion control. Since NoC is implemented on a chip, it is limited by area and power consumption. Choose a simple ...

AI Technical Summary

Problems solved by technology

[0014] 1) The source node needs to occupy the landing platform for local-to-local messages, but because the upstream node continuously sends remote-to-local messages to the source node, the source node cannot send local-to-local messages for a long time;

[0015] 2) The source node has a local-to-remote message and needs to occupy the forwarding platform, but because the upstream node continuously sends remote-to-remote messages to the source node, the source node cannot send local-to-remote messages for a long time

Images