What parameters can be adjusted to further improve concurrent write performance in TiDB, which currently handles 70,000 writes per second?

Note:
This topic has been translated from a Chinese forum by GPT and might contain errors.

Original topic: tidb目前并发写入7w/s,通过调整哪些参数,可以再提高一下并发写入的性能

| username: vcdog

[TiDB Usage Environment] Production Environment
[TiDB Version] v6.5.0
[Reproduction Path] TiDB currently has a concurrent write rate of 70,000/s. Which parameters can be adjusted to further improve concurrent write performance?
[Encountered Problem: Problem Phenomenon and Impact] In the production environment, the TiDB cluster currently has a concurrent write rate of around 70,000/s, which causes the TiDB instance to crash and restart. Is there any way to adjust certain parameters to further improve concurrent write performance?
[Resource Configuration] Go to TiDB Dashboard - Cluster Info - Hosts and take a screenshot of this page
[Attachments: Screenshots/Logs/Monitoring]

| username: tidb狂热爱好者 | Original post link

Buy a VIP service, and the teacher will teach you how.

| username: tidb狂热爱好者 | Original post link

Can’t rely on getting things for free.

| username: rebelsre | Original post link

Horizontal scaling of tidb-server…

| username: magic | Original post link

The information is a bit limited. Please first provide your parameters, the process of your stress testing, and which parameters you have tried to optimize.

| username: vcdog | Original post link

I want to buy it, but the company doesn’t want to, haha.

| username: TiDBer_Qt | Original post link

There is too little information. Please provide the monitoring information. Is it the KV that is down or the TiDB that is down?

| username: vcdog | Original post link

Currently, this TiDB cluster has 3 TiDB-server nodes, with load balancing done through F5. The table structure has only 4 or 5 fields, all of which are relatively short.

| username: vcdog | Original post link

This was the concurrent write during the stress test conducted several months ago, and the monitoring data is no longer available.

| username: 托马斯滑板鞋 | Original post link

:upside_down_face: Upgrading to 7.5 can improve insert performance by about 40%
P.S: Just checked the latest performance test report, and 6.5.6 is actually the highest??? (Not sure what was updated, but when 7.5 first came out, there was a test report showing it had 50% higher insert performance than 7.1 :joy:)

| username: miya | Original post link

The information is indeed quite limited. Since the stress test caused a crash, it indicates that the service has reached its maximum capacity for handling requests. You can improve insertion efficiency by horizontally scaling TiDB and TiKV, and setting the primary key to a random method to avoid primary key generation hotspots.

| username: DBAER | Original post link

First, vertically upgrade the hardware resources, then horizontally scale the linear resources, such as TiDB and TiKV.

| username: zhanggame1 | Original post link

Are the three TiDB servers and TiKV deployed together? How could they crash?

| username: tidb菜鸟一只 | Original post link

Please provide the topology structure. Also, do you experience any hotspots during insertion? How is the TiKV pressure?

| username: Hacker_PtIIxHC1 | Original post link

Expand TiDB and TiKV nodes

| username: xiaoqiao | Original post link

Writing to multiple TiDB nodes can improve concurrency.

| username: xiaoqiao | Original post link

What tool did you use? What is the scenario?

| username: 像风一样的男子 | Original post link

When it crashes under load, which bottleneck is it: CPU, memory, or disk I/O? Whichever is lacking, add more of that.

| username: zhang_2023 | Original post link

Expand the nodes.

| username: 友利奈绪 | Original post link

Upgrade version, expand cluster, upgrade configuration.