Multicore processor based flow control method and multicore processor

Abstract

The embodiment of the invention provides a multicore processor based flow control method and a multicore processor. In the method, a first core in the multicore processor carries out priority sorting on received messages, and sends the sorted messages to at least one middle core which is other single core except the first core in the multicore processor; and any middle core in the at lest one middle core receives the messages sent by the former single core and forwards the messages. Compared with the prior art, the message priority sorting provided by the embodiment of the invention can not only effectively prevent the congestion in mirconetworks, but also discard the message with lower priority to ensure the message with higher priority during the congestion in the micronetworks. Therefore, better QoS (Quality of Service) can be provided in the micronetworks, and further the QoS of macronetworks is further ensured.

Description

technical field [0001] The invention relates to the communication field, in particular to a multi-core processor-based flow control method and the multi-core processor. Background technique [0002] Because the network is always composed of limited resources, including information buffers in network nodes, node processors and input / output links, etc. If no control measures are added to the data transmission, there may be uneven data flow in the network. The data on some nodes and links exceeds the storage and processing capabilities of the nodes, or exceeds the transmission capacity of the links, resulting in network congestion. . Typically, node devices at the edge of the network act as forwarding units and are responsible for performing flow / congestion control. The cores of these forwarding units include network processors, multi-core processors and programmable gate arrays. With the increasing network traffic, processors are gradually developing towards multi-core, and...

AI Technical Summary

Problems solved by technology

[0003] Similar to the macro-network composed of node devices, in order to ensure that the packets in the micro-network are forwarded in an orderly manner, it is usually necessary to take some measures to preserve the order of messages in the micro-network, and these measures to preserve the order of messages It also constitutes the cause of the traffic burst in the micro-network, causing the micro-network to be congested before the data flow is forwarded.

[0004] For example, the multi-thread processing mechanism of a multi-core processor can easily cause the packets of the same data stream to be out of order. For this, the usual way to preserve the order is to use the pipeline method between multiple threads to transmit the starting semaphore, and the thread that gets the semaphore executes Processing, and the processing time of threads is sometimes uneven, resulting in traffic bursts, and then congestion of the micro network

For another example, the front-end single core in a multi-core processor increases the sequence number of the message when processing the message, and the back-end single core needs to reorder the message cache to achieve message order preservation. This method of message order preservation will also Cause traffic bursts, and then micro-network congestion

[0005] Since the existing technology processes the messages transmitted in the micro-network, the forwarding of the messages between the single cores only adopts the simple forwarding mode of FIFO (First In First Out), and each single core constituting the multi-core processor The performance is not completely consistent. For example, the performance of the single core at the front end may be higher than that of the single core at the back end. In this way, once the traffic burst in the micro network leads to congestion, and then the packet needs to be discarded, the packet The simple forwarding mode of FIFO is easy to cause high priority packets to be discarded

Therefore, even though quality of service (QoS, Quality of Service) control is performed at the sending node of the macro network, the above-mentioned abnormality occurring inside the multi-core processor (in the micro network) has actually destroyed the QoS of the network

Images