Be careful with my findings, I appear to have compile in debug mode, I am redoing the benchmarksI recently had to work with many customers having large number of connections opened in MySQL and although I told them this was not optimal, I had no solid arguments to present. More than once, I heard: it is not a problem, those connections are Sleeping! In order to remedy to this situation, I decided to run a series of DBT2 benchmarks while the number of idle connections was varied from 0 to 1500. I was expecting an influence, because those idle threads are presents in internal MySQL list objects that need to be scanned and also the socket SELECT call needs to build a FD bitmap after every network activity and the size of the bitmap is proportional to the number of active threads. What I found is not a small impact at all.
For my daily work, I use 2 computers linked with a gigabit switch so I decided to use them for benchmarking. On the desktop, I installed Percona-Server-11.2 which I configured with the following Innodb settings: innodb_buffer_pool_size=1G
innodb_log_file_size=64M
innodb_log_buffer_size=8M
innodb_flush_method=O_DIRECT
innodb_flush_log_at_trx_commit=0
max_connections = 10000
open_files_limit = 32768
Read more: MySQL Performance Blog
For my daily work, I use 2 computers linked with a gigabit switch so I decided to use them for benchmarking. On the desktop, I installed Percona-Server-11.2 which I configured with the following Innodb settings: innodb_buffer_pool_size=1G
innodb_log_file_size=64M
innodb_log_buffer_size=8M
innodb_flush_method=O_DIRECT
innodb_flush_log_at_trx_commit=0
max_connections = 10000
open_files_limit = 32768
Read more: MySQL Performance Blog
0 comments:
Post a Comment