maria-developers team mailing list archive

[Axel Schwenke <axel@xxxxxxxxxxxx>]

Hi,

Monty asked me to do a quick check how MariaDB performs with jemalloc. This
is in the light of the upcoming integration of TokuDB into the MariaDB tree.
TokuDB requires jemalloc.

While I did use tcmalloc in the connect speed benchmarks with good results,
I did no use jemalloc before. This article from Percona

http://www.mysqlperformanceblog.com/2013/03/08/mysql-performance-impact-of-memory-allocators-part-2/

suggests that jemalloc and tcmalloc perform very comparable, but also that
jemalloc shows a regression in 3.2.0 and later.

The benchmark I used is sysbench OLTP ro, 16 tables, data fitting into
buffer pool. I used the new 64 (virtual, 32 real) core machine and compared
4 different malloc implementations as below:

OLTP ro transactions per second relative to glibc malloc
----------------------------------------------------------------------
Threads	jemalloc-3.3.1	glibc malloc	tcmalloc	jemalloc-3.1.0
4	+3.2%		+0.0%		+11.6%		+61.9%
8	+0.9%		+0.0%		-2.3%		+8.0%
16	+9.6%		+0.0%		+4.7%		+12.9%
32	+1.8%		+0.0%		+0.9%		+5.8%
64	-5.1%		+0.0%		+2.1%		+5.4%
128	-4.8%		+0.0%		+0.2%		+4.5%
256	-4.2%		+0.0%		+0.7%		+5.0%
512	-13.3%		+0.0%		+1.8%		+4.1%

(more data in attached spread sheet)


Observations:

- the effect is rather limited
- jemalloc 3.1.0 clearly wins
- jemalloc 3.3.1 performs bad at higher concurrency


Conclusion

It's fine to build MariaDB with jemalloc in the future. However we mustn't
use jemalloc-3.3.1, but rather jemalloc-3.1.0.


XL

Attachment: OLTP-malloc-series40.ods
Description: application/vnd.oasis.opendocument.spreadsheet