<<toc>>
Here are some benchmarks of some development threadpool code.
The benchmarks were run on three machines:
* facebook-maria1 (Linux, 16 cores)
* pitbull (Linux, 24 cores)
* windows (Windows, 8 cores)
Sysbench 0.4 was used to run some "unit" OLTP tests (point-select and
update-nokey), as well as the "classic" OLTP-readonly and OLTP-readwrite. All
tests were run with the number of concurrent clients ranging from 16 to 4096,
with warm cache, with the sysbench table having 1M records.
The results are quite different on all of the machines tested (the machines are
very different, in terms of cores, IO etc), yet threadpool has a positive
effect on all 3 machines, and the positive effect seems to grow with the number
of cores.
Some notes on how the benchmarks were run:
# The benchmark client and server used different CPUs - ('taskset'
(Linux), or 'start /affinity' (Windows) were used to run the benchmark client
on <<code>>#CPUs/4<</code>>, the rest of CPUs were used by the server). On
the Linux boxes, <<code>>--thread_pool_size=<N><</code>> (where N is number
of cores dedicated to the server) was used.
# ##innodb_flush_log_at_trx_commit=0## and ##innodb_flush_method=ALL_O_DIRECT##
was used to avoid extensive fsyncing, which is ok for the purposes of the
testing for this.
# Every "write" benchmark (##oltp_rw## and ##update_nokey##) started with a new
server (i.e. kill mysqld, remove innodb files, and restart mysqld for each
test). Every "read" benchmark, on the other hand, reused the same running
server instance. Starting afresh with a new server on write benchmarks is
done mainly to eliminate the effects of the purge lag.
== OLTP_RO
=== OLTP_RO facebook-maria1
<<style class="darkheader-nospace-borders-side">>
|= concurrent clients |16|32|64|128|256|512|1024|2048|4096
|= thread per connection|3944|4725|4878|4863|4732|4554|4345|4103|1670
|= threadpool|3822|4955|4991|5017|4908|4716|4610|4307|2962
<</style>>
{{oltp-ro-facebook-maria1}}
=== OLTP_RO pitbull
<<style class="darkheader-nospace-borders-side">>
|= concurrent clients |16|32|64|128|256|512|1024|2048|4096
|= thread per connection|6754|7905|8152|7948|7924|7587|5313|3827|208
|= threadpool|6566|7725|8108|8079|7976|7793|7429|6523|4456
<</style>>
{{oltp_ro-pitbull}}
=== OLTP_RO Windows
<<style class="darkheader-nospace-borders-side">>
|= concurrent clients |16|32|64|128|256|512|1024|2048|4096
|= thread per connection|1822|1831|1825|1829|1816|1879|1866|1783|987
|= threadpool|2019|2049|2024|1992|1924|1897|1855|1825|1403
<</style>>
{{oltp_ro-windows}}
== OLTP_RW
=== OLTP_RW facebook-maria1
<<style class="darkheader-nospace-borders-side">>
|= concurrent clients |16|32|64|128|256|512|1024|2048|4096
|= thread per connection|2833|3510|3545|3420|3259|2818|1788|820|113
|= threadpool|3163|3590|3498|3459|3354|3117|2190|1064|506
<</style>>
{{oltp_rw-facebook-maria1}}
=== OLTP_RW pitbull
<<style class="darkheader-nospace-borders-side">>
|= concurrent clients |16|32|64|128|256|512|1024|2048|4096
|= thread per connection|4561|5316|5332|3512|2874|2476|1380|265|53
|= threadpool|4504|5382|5694|5567|5302|4514|2548|1186|484
<</style>>
{{oltp_rw-pitbull}}
=== OLTP_RW Windows
<<style class="darkheader-nospace-borders-side">>
|= concurrent clients |16|32|64|128|256|512|1024|2048|4096
|= thread per connection|1480|1498|1472|1477|1456|1371|731|328|82
|= threadpool|1449|1523|1527|1492|1443|1409|1365|1240|862
<</style>>
{{oltp_rw-windows}}
== POINT_SELECT
=== POINT_SELECT facebook-maria1
<<style class="darkheader-nospace-borders-side">>
|= concurrent clients |16|32|64|128|256|512|1024|2048|4096
|= thread per connection|91322|113116|115418|114484|111169|104612|26902|12843|5038
|= threadpool|100359|115618|118115|120136|119165|113931|110787|109970|104985
<</style>>
{{point_select-facebook-maria1}}
=== POINT_SELECT pitbull
<<style class="darkheader-nospace-borders-side">>
|= concurrent clients |16|32|64|128|256|512|1024|2048|4096
|= thread per connection|148673|161547|169747|172083|69036|42041|21775|4368|282
|= threadpool|143222|167069|167270|165977|164983|158410|148690|147107|143934
<</style>>
{{point_select-pitbull}}
=== POINT_SELECT Windows
<<style class="darkheader-nospace-borders-side">>
|= concurrent clients |16|32|64|128|256|512|1024|2048|4096
|= thread per connection|39734|42885|44448|44478|41720|38196|36844|35404|23306
|= threadpool|42143|45679|47066|47753|46720|44215|43677|43093|44364
<</style>>
{{point_select-windows}}
== UPDATE_NOKEY
=== UPDATE_NOKEY facebook-maria1
<<style class="darkheader-nospace-borders-side">>
|= concurrent clients |16|32|64|128|256|512|1024|2048|4096
|= thread per connection|60165|65761|67727|57232|47612|26341|8981|3265|389
|= threadpool|65092|68683|67053|64141|64815|63047|63346|63638|62843
<</style>>
{{update_nokey-facebook-maria1}}
=== UPDATE_NOKEY pitbull
<<style class="darkheader-nospace-borders-side">>
|= concurrent clients |16|32|64|128|256|512|1024|2048|4096
|= thread per connection|65213|71680|19418|13008|11155|8742|5645|635|332
|= threadpool|64902|70236|70037|68926|69930|69929|67099|62376|17766
<</style>>
{{update_nokey-pitbull}}
=== UPDATE_NOKEY Windows
<<style class="darkheader-nospace-borders-side">>
|= concurrent clients |16|32|64|128|256|512|1024|2048|4096
|= thread per connection|24790|25634|25639|25309|24754|19420|5249|2361|824
|= threadpool|25251|25259|25406|25327|24850|23818|23137|23003|22047
<</style>>
{{update_nokey-windows}}
