Note:
This topic has been translated from a Chinese forum by GPT and might contain errors.
Original topic: tidb与MySQL性能差距
[Test Environment for TiDB]
Recently, I have been researching TiDB and configured a three-node TiDB cluster. When using sysbench for bulk insert testing, I found that the performance of the TiDB cluster is significantly lower than that of a single-node MySQL, even though the machine configurations are the same. However, TiDB is set up on three machines, each with PD, TiDB, and TiKV.
Test command:
sysbench bulk_insert --threads=16 --time=60 --report-interval=5 --mysql-host=*** --mysql-port=4000 --mysql-user=root --mysql-password=****** --mysql-db=sbtest --tables=10 --table-size=1000000 run
TiDB test results:
MySQL test results:
I would like to ask what is causing such a large performance gap, otherwise, I won’t be able to explain it later
How is the resource configuration of TiDB compared to MySQL?
The machines are all 8 Core + 16G + 100G.
When I tested it before, it was also a mixed deployment, and the configuration was similar to yours. MySQL also had slightly better performance, but not as exaggerated as yours.
Personally, I think we should compare from a distributed perspective. According to the production configuration, deploy pd3, tidb2, tikv*3 respectively, and then test the performance.
After all, the product architecture design is different.
That’s normal. My tests also yielded similar results. Simply having an empty database won’t make writes faster than MySQL.
The application scenarios are different, and the tracks are different, so there’s no need for a forced comparison.
That’s normal, the network overhead alone is more than doubled; you can also test point_select, I found the difference to be about 3 times;
However, when it comes to multi-table joins in tpcc, it’s the other way around 
If the data volume is not large enough, a single machine definitely has an advantage, as it avoids a lot of distributed overhead.
However, I created a single-node test environment using the playground, and the test results still showed no difference.
It’s normal and related to the underlying storage structure. When I tested the playground on Alibaba Cloud, it was even 2% slower than the cluster. 
You can learn about the multi-layer structure of TiDB’s LSM tree, and even compare it with OB’s two-layer structure + B-tree.
Try a different approach and give it another shot
7 to 8 times higher is too outrageous.
I can only tell that your boss thinks highly of TiDB. Have you considered asking your boss to pay for you to get a certification? Take advantage of this great opportunity.
The leader has already tested it for you.
The leader is very optimistic about TiDB 
|
username: TiDBer_djgos04V
|
username: 
|
username: 
|
username: 
|
username: 
|
username: 
|
username: 
|
username: 
|
username: 
|
username: 
|
username: