Bug #50146 mysql-5.5.X show mixed performance results compared to mysql-5.1.40sp1
Submitted: 7 Jan 2010 17:07 Modified: 29 Oct 2010 15:51
Reporter: Jonathan Miller Email Updates:
Status: No Feedback Impact on me:
None 
Category:MySQL Server: InnoDB Plugin storage engine Severity:S5 (Performance)
Version:5.5.1-m2 & 5.5.99-m3 OS:Linux
Assigned to: Assigned Account CPU Architecture:Any

[7 Jan 2010 17:07] Jonathan Miller
Description:
Automated Performance Testing is showing a mixed set of results in the mysql-5.5.X clones compared to mysql-5.1.40sp1 release.

Sysbench OLTP_RW:

performance was gained in the thread counts above 16 for both 5.5.1-m2 and
5.5.99. Yet with the gain we have a loss of (-12 % ) on the lower thread counts below 16. 

Sysbench SELECT_SUM_RANGES:

we see a good gain with the lower thread count (8,16), yet we start a sharp decline around 64 threads. (64 = -4 %, 128 = -10 %, 256 = -8 %) for 5.5.1-m2 and 5.5.99.

Sysbench SELECT_SIMPLE_RANGES:

Simple Ranges shows many of the same attributes as Sum Ranges. 5.5.99 shows better gains in
the lower thread count and less of a drop in the higher threads then does 5.5.1-m2.
5.5.1-m2 – (64 = -6 % , 128 = -10 %, 256 = -6 % )

How to repeat:
There are 2 ways one can repeat these tests.

1) Use AB2

Pull autobench2 from the mysql-test-extra clone change the location settings in the ab-default.env file

Commands to start testing:

/bin/bash <location of ab2>/ab2_boot.sh --clone=mysql-5.1 --revtag=mysql-5.1.40sp1 --conf=<location of ab2>/ab-default.env --test=sysbench_innodb

/bin/bash <location of ab2>/ab2_boot.sh --clone=mysql-trunk --conf=<location of ab2>/ab-default.env --test=sysbench_innodb

/bin/bash <location of ab2>/ab2_boot.sh --clone=mysql-next-mr --conf=<location of ab2>/ab-default.env --test=sysbench_innodb

2) Run by hand

Pull desired clone

use the compile-pentium-max build script.

Once installed, start mysqld using the configurations found in autobench2/conf/sysbench-std/sysbench-std-innodb.cnf

Run sysbench test by hand.

Note: ab2 runs 3 iterations of each thread and then averages the results to ensure stability.
[29 Oct 2010 23:00] Bugs System
No feedback was provided for this bug for over a month, so it is
being suspended automatically. If you are able to provide the
information that was originally requested, please do so and change
the status of the bug back to "Open".