Scenes
Database stress testers found that statistical queries were very slow.
Enter the service to check free, and find that the cache/buffer is very large.
implement
echo 3 > /proc/sys/vm/drop_caches
cache/buffer is normal.
But the query is still slow
See the size of mysql's innodb_buffer_poll, which is smaller than the file size. Causes a lot of files to be cached by the operating system.
Increase innodb_buffer_poll size, ok
Just talk about disk read and write
The HDD disk given by the company, and it is still 7200 rpm, is dead.
Then the iops per second is 7200/60=120.
innodb_page size is 16K
The random read and write per second is 120*16 = 1920K, and the maximum is 2M. 7.2G per hour
24G data, it takes more than 3 hours to read.
It only took 573 seconds to load 24G of data into memory.
Then the sequential read per second is 573 seconds
24*1024*/573 = 43M/S
This speed is still unacceptable.