percona-discussion team mailing list archive

[MARK CALLAGHAN <mdcallag@xxxxxxxxx>]

These are sysbench readonly results from a 16 core server. I used my
--oltp-secondary-index patch. The workload is CPU bound. The numbers
are transactions/second for 1, 2, 4, 8, 16, 32 and 64 concurrent
threads. With my sysbench patch, 2X more work is done per transaction
so the results should be ~1/2 of what you get. tcmalloc helps the
binaries in which other SMP problems have been fixed. I have not
debugged the slowdowns for the other cases. Postgres thrashes for 64
concurrent users, so I don't include that result.

 311  684 1305 2401 2678 2211    0    pg
 213  460  874 1361  836  732  522     5075p
 210  468  841 1260  713  561  490     5075ptcm
 209  473  900 1432 1766 1102 1069   5077percb13hp
 209  477  879 1477 2015 2070 2019   5077percb13hptcm
 192  417  755 1182 1245  733  621    5131perc
 198  435  796 1256 1573 1382 1118   5131perctcm

Legend:
- pg -- Postgres 8.3.6
- 5075p -- Percona build 12
- 5075ptcm -- Percona build 12 + tcmalloc
- 5077percb13hp -- Percona high-perf build 13
- 5077percb13hptcm -- 5077percb13hp + tcmalloc
- 5131perc -- Percona for 5.1.31
- 5131perctcm -- 5131perc + tcmalloc


-- 
Mark Callaghan
mdcallag@xxxxxxxxx