A memory allocation method and serv

Abstract

The embodiment of the present application discloses a memory allocation method and a server, which are used for reducing performance loss caused by NC delay and improving server performance when memory allocation is performed. A method for embodiment of that present application includes: Server Identification Node Topology Table, The node topology table contains not only the connection relationships between NUMA nodes, but also between the NUMA node and the NC, the connection between NCs, Based on the node topology table, generating a memory access jump table of each NUMA node, The hop table contains not only the number of QPI hops in the shortest path connected to other NUMA nodes, and has an NC hop count, according to the access jump table of each NUMA node, calculating the memory accesspriority of each NUMA node, The number of NC hops is taken as an important parameter in the calculation of memory access priority. The less the number of NC hops, the higher the memory access priority. When a NUMA node applies for memory, the memory is allocated according to the memory access priority table. The higher the priority, the more priority the memory is allocated from the NUMA node corresponding to the priority.

Description

technical field [0001] The application relates to the communication field, and in particular to a memory allocation method and a server. Background technique [0002] Limited by hardware chip technology, the computing power of a single processor (Central Processing Unit, CPU) tends to be saturated. Therefore, in order to obtain higher computing performance, the server tends to increase the computing performance of the server by increasing the number of processors. For high-performance servers, Non-Uniform Memory Architecture (NUMA) is generally used, that is, multiple nodes are connected through a high-speed Internet, and each node is composed of a group of CPUs and local memory. When the node accesses the local memory, the memory access delay is small and the performance is high; however, if the node accesses the remote memory, the memory access delay is relatively high, which will cause performance degradation. Therefore, in order to improve system performance, the curre...

AI Technical Summary

Problems solved by technology

This leads to different transmission overheads between different NUMA nodes

Therefore, the calculation method of NUMA distance in the above-mentioned memory allocation method cannot realize the problem of NC, which will lead to an increase of cross-NC memory access, which will increase memory access delay and lead to server performance degradation

Images