NUMA 架构下,是否还应限制单服务使用的内存
Open
@shichun-0415 is already working on this.
Since Jul 14, 2022.
area/deploy-upgrade-maintain
- Dominant language
- Python
- Stars
- 1.8k
- Forks
- 1.2k
- Avg merge
- 1d 16h
- Merged PRs (30d)
- 31
Description
Error 报告
This repository is ONLY used to solve problems related to DOCS-CN.
For other issues (TiDB, TiKV, etc), please move to other repositories.
Please answer the following questions before submitting your issue. Thanks!
- What is the document directory for the problem?
- Github: https://github.com/pingcap/docs-cn/blob/master/benchmark/benchmark-tidb-using-sysbench.md
- 文档站:https://docs.pingcap.com/zh/tidb/stable/benchmark-tidb-using-sysbench#%E5%9C%A8%E9%AB%98%E5%B9%B6%E5%8F%91%E5%8E%8B%E5%8A%9B%E4%B8%8Btikv-%E4%B9%9F%E6%9C%AA%E8%BE%BE%E5%88%B0-cpu-%E4%BD%BF%E7%94%A8%E7%93%B6%E9%A2%88%E4%B8%BA%E4%BB%80%E4%B9%88-tidb-%E7%9A%84-cpu-%E5%88%A9%E7%94%A8%E7%8E%87%E4%BE%9D%E7%84%B6%E5%BE%88%E4%BD%8E
- How would you like to improve it?
原文:

此处提到了 NUMA 架构,但未限制单进程的内存用量,这有可能会导致跨 NUMA Node 访问内存。跨 NUMA Node 访问内存,会极大地降低性能(早期服务器 7 倍,后期 1.3 倍)。
仅限制了:
进程数=NUMA CPU 的个数(其实此处是否应该使用NUMA Node 数或物理 CPU 数替代NUMA CPU 的个数来消除歧义?)单个 NUMA Node 中 CPU 的核心数=max-procs
那么在使用过程中,肯定会有部分 TiDB 申请额外的 NUMA Node 外的内存,因为进程并非绝对平分内存。应额外添加:
单个 TiDB 进程可使用的内存上限=单个 NUMA Node 内的内存容量(即:单路内存容量)
限制进程内存使用。
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Assessment
This issue has not been assessed yet.