A MPI application acceleration system and method based on photoelectric hybrid switching network

Abstract

The invention discloses an MPI application acceleration system and method based on a photoelectric hybrid switching network. The system of the invention includes a computing node, a top-of-rack switch, an optical path switch, a core switch and an MPI scheduling controller. The top-of-rack switch is connected to the optical path switch, the core switch and the MPI scheduling controller respectively, and the MPI scheduling controller is connected to the optical path switch at the same time. The method of the present invention can prioritize different communication operations in MPI applications, and generate data traffic allocation weights based on priority and data volume, and the top-of-rack switch sends data traffic to the optical path switch for acceleration according to the weight value, greatly Improve bandwidth resource utilization.

Description

technical field [0001] The invention belongs to the technical field of computers, and in particular relates to an MPI application acceleration system and method based on a photoelectric hybrid switching network. Background technique [0002] MPI (Message Passing Interface) is a parallel programming technology based on information passing, which is widely used in high-performance computing parallel programming model. With the continuous improvement of computing power, the node scale of high-performance computers is increasing, and it is expected to reach hundreds of thousands of nodes in the next few years. The scale of various applications running on high-performance computers continues to increase, the complexity of the programming model of parallel applications continues to increase, and the execution efficiency and completion time of parallel applications are challenged. When the MPI-based high-performance computing parallel programming model reaches thousands to tens of...

AI Technical Summary

Problems solved by technology

However, there are also problems in the optical interconnection network: first, the cost of network equipment such as optical transmission modules and optical switches is high, and the cost of large-scale deployment of equipment is huge; second, the granularity of service traffic support is not flexible enough, and some services with small data volumes Traffic has a low utilization rate of bandwidth resources

Therefore, the scheduling of traffic in the network is very important and directly affects network performance. However, this paper distinguishes data traffic at the application layer with a large granularity. The distinguishing standard is only the size of the data, and it is impossible to prioritize data traffic according to specific application operations. distinguish

Images