[MDEV-17261] sysbench oltp read only too slow for MyRocks. - Jira

XML

Word

Printable

(Background: So I was directed to this performance comparison result:
https://twitter.com/thewebscaledba/status/1002604510991142912?s=12 They report MyRocks-in-MariaDB is slower than InnoDB-in-MySQL)

This is oltp-read-only, the data is on OS cache. CPU-bound read-only workloads are not a MyRocks strong point.

Still, I got an aws ec2 c5.4xlarge (16 VCPU, 32G RAM) and ran tests on current MariaDB 10.3. Non-default settings:

log_bin=1

sync_binlog=1

binlog_format=row

max_connections=500

sysbench /usr/share/sysbench/oltp_point_select.lua --table-size=5000000

--threads=1 --rand-type=uniform --db-driver=mysql --mysql-socket=/tmp/mysql.sock

--mysql-user=root --mysql_storage_engine=$engine    prepare

sysbench /usr/share/sysbench/oltp_point_select.lua --table-size=5000000 --threads=$threads

--time=180 --rand-type=uniform --db-driver=mysql --mysql-socket=/tmp/mysql.sock

--mysql-user=root run

The result is not good:

InnoDB:

n_threads   TPS (QPS is the same)

100 181,984

150 182,268

200 178,739

RocksDB

n_threads   TPS (QPS is the same)

100 7,771

150 5,979

200 4,856

The difference is huge!

relates to

MDEV-13845 benchmark RocksDB engine

MDEV-18080 Run MyRocks benchmark: MariaDB vs Percona Server vs FB/MySQL

Error rendering 'com.xiplink.jira.git.jira_git_plugin:git-issue-webpanel'. Please contact your Jira administrators.