[Bug]: sysbench performance: same table size, 1000 threads have lower QPS than 100 threads or not higher，but tidb shows linear growth

heni02 commented 1 year ago

- Version or commit-id (e.g. v0.1.0 or 8b23a93):nightly-fd2af5d
- Hardware parameters:
- OS type:
- Others:

Currently, mo sysbench performance found 1000 threads have lower QPS than 100 threads ，whatever select/update/delete/insert test ,but tidb shows linear growth last month performace data： https://doc.weixin.qq.com/sheet/e3_AREAuQb9AC8tOXj19g4SVOMsRCO2H?scode=AJsA6gc3AA8fVeui2vAYYAgwazACs

No response

No response

No response

nnsgmsone commented 1 year ago

远宁安排个人把。。我要优先处理mpool和1240/1241这三个s-1 @ouyuanning

ouyuanning commented 1 year ago

@heni02 是否跟之前测试tpcc的问题是一样的。

ouyuanning commented 1 year ago

跟CN提供并发的能力有关。感觉属于性能优化范畴

LiSong0214 commented 1 year ago

单机并发性能不符合预期，请 @XuPeng-SH 看看

XuPeng-SH commented 11 months ago

1000 单机比 100 差一些概率是我们的单机性能饱和，加上十倍的并发带来的overhead, 具体要在1.2后续分析

XuPeng-SH commented 6 months ago

@volgariver6 given the same 3 CNs, I think the big difference between MO and TIDB is the data cache

volgariver6 commented 3 months ago

未处理

volgariver6 commented 2 months ago

未处理

volgariver6 commented 2 months ago

未处理

volgariver6 commented 2 months ago

未处理

volgariver6 commented 2 months ago

未处理