percona-discussion team mailing list archive

[MARK CALLAGHAN <mdcallag@xxxxxxxxx>]

I ran sysbench with my changes to make it use predicates on a
secondary index for half of the queries. Vadim has the diff for the
changes to sysbench.

I used:
* 8 core server
* CPU bound workload
* 1,2,4,8,16 concurrent sessions
* Xtradb 5.0.75 build 12 compiled from source --
http://www.mysqlperformanceblog.com/2009/01/23/5075-build12-percona-binaries/
* with and without tcmalloc
* throughput is value reported as 'transactions:'
* sysbench binary runs on same host as db server
* sysbench command line is: (--oltp-secondary-index is new with my patch)
  --test=oltp --oltp-read-only --oltp-table-size=2000000
--max-time=120 --max-requests=0 --mysql-table-engine=innodb \
  --db-ps-mode=disable --mysql-engine-trx=yes --oltp-secondary-index

Results:

 throughput for      1     2     4     8    16    concurrent sessions
without-tcmalloc  220 413 718 461 397
with-tcmalloc      223 439 744 780 759

Observations:
The only thing needed to make CPU-bound sysbench readonly fast is
tcmalloc. MySQL performance doesn't change much with Percona or Google
patches compared to the original code. This may change when the
bottlenecks above InnoDB (LOCK_open) are removed and when run on a
larger box.

---
Mark Callaghan
mdcallag@xxxxxxxxx