fangel's photostream  

MySQL Benchmark

I wanted to test if Master-Slave was indeed faster, better & stronger.

However, I didn't have three computers around, so I installed three Virtual Machines (Host and VMs all OpenSuse 11.0).
This, however, affected the results, because when I ran the master-only tests, the slave wasn't doing anything, thus releasing resources that the web- and master-db-server could use.

For this to give real results, I would either need three dedicated servers (anyone wanna sponsor?) or find a way to limit each VM to a specific amount of CPU, so they can't "borrow" from the inactive slave in the master-only tests.

The tests did the following:
1) Connect to the db(s). So for the master-only tests, one connection. For the master-slave tests, two connections.
2) SELECT * FROM table1 (three entries)
3) INSERT INTO table1 (field, field, field) VALUES (value, value, value)
4) SELECT * FROM table1 WHERE id=[just-inserted]
5) DELETE FROM table1 WHERE id=[just-inserted]
6) SELECT * FROM table2 (three entries)

The Y-axis is milliseconds for the complete request (setup, wait & transfer)
The X-axis is number of concurrent connections.

All tests, ran for 10.000 requests.

I used AB from a seperate computer to create the requests.. Plotted time is "Time per request (mean, across all concurrent requests)"

Some rights reserved

Uploaded on Oct 12, 2008

0 comments