Performance Testing Issues with TiDB Using Sysbench

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

Original topic: tidb性能使用sysbench性能测试问题

| username: 气死人的萌新

[TiDB Usage Environment] Test environment | Standardized cluster deployment according to the official documentation
[TiDB Version] V6.5.4
[Reproduction Path] Followed the steps in this document How to Test TiDB with Sysbench | PingCAP Documentation Center and this document TiDB Cloud Sysbench Performance Test Report | PingCAP Docs for testing.
[Encountered Problem: Problem Phenomenon and Impact]
Sysbench configuration file
image
Tested with threads switched to 50 and 100 respectively. Below are the test statements and results



image
Looking at the Sysbench performance in the test environment of v7.1.0, the difference is too much, not just this test, but several other tests also differ greatly. Is there a configuration issue on my side? Please advise.
image

[Resource Configuration]
image
image

[Attachments: Screenshots/Logs/Monitoring]

| username: 气死人的萌新 | Original post link

Here are my other result analyses:

| username: 气死人的萌新 | Original post link

The image you provided is not accessible, so I cannot translate its content. Please provide the text you need translated.

| username: 气死人的萌新 | Original post link

Planning to test with 200 concurrent users later.

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

Is the hard drive an SSD? The read and write performance of the database is highly related to the performance of the disk.

| username: 气死人的萌新 | Original post link

Using a virtual hard disk

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

If it’s a virtual machine, the disk performance is shared, so your stress test results won’t look good. The official recommendation for hard drives is SSD, preferably dedicated NVMe disks.

| username: 气死人的萌新 | Original post link

I just asked, the disk is using SSD, SATA. However, because it is a virtual disk, there should be some performance loss.

| username: TiDBer_小阿飞 | Original post link

Does it have anything to do with the operating system version? Red Hat 8 doesn’t seem very friendly.

| username: 气死人的萌新 | Original post link

It should be fine according to the official documentation, right? Red Hat 8

| username: 气死人的萌新 | Original post link

The image you provided is not accessible. Please provide the text you need translated.

| username: 气死人的萌新 | Original post link

I just don’t know if this performance loss is normal. The difference is quite large compared to 7.1 TiDB Cloud System Performance Test Report | PingCAP Documentation (pingcap.com).

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

You can increase the concurrency to 500 or 1000, and the results will look much better.

| username: 有猫万事足 | Original post link

There is a big difference between SATA and NVMe.

You can take a look at this.

| username: 气死人的萌新 | Original post link

Bro, according to what you just said, I changed the concurrency to 500, and here is the data I got:
image
image

| username: 气死人的萌新 | Original post link

This is the data for 100 concurrent users.

| username: 气死人的萌新 | Original post link

It has decreased instead of increasing. Do you have any suggestions?

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

This is the result of my previous test, for reference only.
image

| username: 大飞哥online | Original post link

If it’s a virtual machine, check if it’s sharing the host machine. During stress testing, first identify where the bottleneck of the machine is.

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

During the stress test, output the dstat of a TiKV host

dstat 1
# If not installed, install it
yum install dstat