I'm currently doing some trials of Galera Cluster with Percona 5.6. I've progressed to testing a 2 node cluster on baremetal. With repl.commit_order
set to 0 (off) the TPS performance as measured with sysbench is comparable to that of a standalone Percona instance.
When repl.commit_order
is set to 3 (the default), TPS performance is around 10% compared to that of a standalone Percona instance.
Sysbench command in each instance is:
sysbench --test=/usr/share/doc/sysbench/tests/db/oltp.lua --oltp-table-size=5000000 --oltp-tables-count=1 --num-threads=32 --mysql-host=127.0.0.1 --mysql-user=root --mysql-password=**** --mysql-port=3306 --report-interval=1 --max-requests=0 --max-time=0 --mysql-db=sbtest run
Firstly: What are the implications of the different commit_order modes?
Secondly: Can anyone suggest any other tweaks that will improve performance?