Selective explicit congestion notification marking method for data center

Abstract

The invention discloses a selective explicit congestion notification marking method (SECN for short) for a data center; the SECN implementation is divided into two parts: 1, a sending end writes estimated flow rate into the head of a data packet; 2, a switch judges the state of the flow speed rate and the like by reading the rate value of the head of the data packet, and the switch marks the flowwith high sending speed rate. The method, namely the SECN, in the invention is a first ECN marking method for distinguishing the flow state, and the method is low-load. According to the method, the link throughput can be obviously improved, the flow finishing time can be shortened, and the throughput of the multi-bottleneck flow can be improved.

Description

technical field [0001] The invention belongs to the technical field of data centers, and more specifically relates to a selective explicit congestion notification marking method for data centers. Background technique [0002] In today's data centers, computer clusters support many network applications, such as web search, social networking, recommendation systems, etc. Users are very sensitive to the response speed of these network applications, requiring computer clusters to complete user application requests within a specified time. [0003] In the data center, these applications are often decomposed into many small tasks, and these small tasks are distributed on different servers in the computer cluster and executed in parallel. These network applications can only be completed through frequent data interaction and mutual cooperation between servers. In this way, the network data transmission performance of computer clusters has a great impact on the response speed of app...

AI Technical Summary

Problems solved by technology

However, there are two shortcomings in this method: first, too low a threshold often triggers a lot of ECN congestion signals, causing the sender to overreact, resulting in a loss of network throughput, which in turn reduces the transmission performance of network data; Second, this ECN marking method based on the instantaneous queue length is an ECN marking method that is agnostic to the flow state. In this method, regardless of the state of the flow where the data packet is located, as long as the data packet exceeds the threshold, it will be marked with an ECN mark. Although the low-speed data flow is not the main cause of congestion, it will be marked with ECN, and the sender will reduce the sending rate. This is very unfair to the low-speed flow and will greatly increase the completion time of the low-speed flow.

To make matters worse, most of the streams in the data center are randomly arriving short streams, these data streams are of short duration, and the sending rate is very low, once these streams are marked with ECN congestion signal, their sending rate will be forced decrease, causing the completion time of these streams to increase significantly, thereby increasing the response time of the network application

Images