MySQL tests server utilization with strong performance

It can be seen that in the case of 1246 instances, tps is basically linear growth. Although there is no absolute adequacy, there may be other considerations in the actual situation.

It can be seen that in the case of 1/2/4/6 instances, tps is basically linear growth. Although there is no absolute adequacy, there may be other considerations in the actual situation.

The usage of MySQL 5.5 to multi-core cpu recorded in the previous article, click here (). In fact, the main purpose of the original test was to say today: what will happen if a server with strong performance runs multiple instances to make full use of the entire machine?

The tested machines are DELL R720XD, 24-core cpu, 96 GB memory, and 24 SAS disks mounted. Before the test, two raid 10 disk arrays were created for the 24 disks, six logical partitions were created for each raid 10 Disk Array Using GPT, And the ext3 File System was created. Test Tool sysbench, test-mode = complex, buffer = 12 GB for each MySQL instance, and about 36 GB of data is initialized during the test. When 1/2/4/6 instances are run separately, the total tps and rw request/s are obtained using the statistical data generated by sysbench. The following is a graph:

It can be seen that in the case of 1/2/4/6 instances, the tps is basically linear growth. Although there cannot be absolute controllability, in actual situations, there may be other factors to consider, but you can also make a reference, coupled with the MySQL 5.5 previously tested for cpu multi-core utilization, as MySQL functions become more and more powerful, it looks like buying a better machine to run multiple instances is more suitable than buying a normal machine to run a single instance on each machine.

In addition, the following figure shows the MySQL 5.1 test result. The overall performance is 5.5-20% lower than that of MySQL ~ About 30%, of course, many OS parameters were not adjusted during the test, and MySQL itself was also not tuned, but the two were tested in the same standard:

From this we can see that the tps is lower than 5.5 in an incomplete linear growth. Haha, find a stable version 5.5 that has been upgraded.