The problem of the MySQL engine

[yubbandzhu]

Hello，
I want to replacing InnoDB engine with tokudb engine, because my enviroment is large, now there are more than 10000+ servers on my zabbix， also the server number in the future will be several times as many times as now. If I use the tokudb, the performence of the zabbix server DB will be improved？ and Will there be any problems in the future？

[kloczek]

I didn't see how the move from innodb to tokudb may solve any of the scale issues.
The number of the servers has no meaning.

Only factor showing enough well scale factor is NVPS + ratio between the number of metrics and triggers.

[yubbandzhu]

Is that the performance of data insertion is better if use tokudb engine? Is it more advantageous than InnoDB when the number of servers is increasing？

[kaspars.mednis]

Hi,

i suggest you first take a look here

https://support.zabbix.com/browse/ZBXNEXT-2363

tokudb is not officially supported by Zabbix at the moment, and you can run into lot of upgrade, indexing and other issues.

The number of hosts does not really matter, the problem is - how much NVPS are you getting ? (New Values per Second)
Have you partitioned your MySQL database ?

Regards,
Kaspars

[yubbandzhu]

Hi,

i suggest you first take a look here

https://support.zabbix.com/browse/ZBXNEXT-2363

tokudb is not officially supported by Zabbix at the moment, and you can run into lot of upgrade, indexing and other issues.

The number of hosts does not really matter, the problem is - how much NVPS are you getting ? (New Values per Second)
Have you partitioned your MySQL database ?

Regards,
Kaspars

The NVPS of my zabbix enviroment is 11695.18 now. There's nothing wrong with it for the time being. Yes, I partitioned my mysql databases.

[kloczek]

Is that the performance of data insertion is better if use tokudb engine? Is it more advantageous than InnoDB when the number of servers is increasing？

It is not possible to solve insertion speed by change engine only.
There are many things outside only storage engine which cannot be solved by the change from innodb to toku.
If you want to improve write speed first you need to lower read IOs latency because before writing something it is necessary to read some number of data to locate the position where new data needs to be written. Key to lower this latency is enough memory to cache those data.
The typical indicator that system has enough memory to cache those data is at least 1:20 ratio between read and write IOs.
If wit at least such ratio between R and W IOs write speed is still not enough at the moment the only solution is witching to experimental MySQL 8.0.x
Full explanation why it is like this you can find in the last FOSDEM presentation https://www.youtube.com/watch?v=gpkuKppwiiE