Brian "Krow" Aker (krow) wrote,
Brian "Krow" Aker

GUID VS Autoincrement, Take Two

One of the projects I am looking at right now is performance on eight
proc (2 proc/4core) machines that Intel is providing.

An interesting situation I've noticed is the behavior in Innodb in
regards to Innodb and primary key INSERTS. From observing behavior
with GUID vs Autoincrement, Innodb runs into issues around ~75
concurrent acting connections (aka not idle). GUID size aside, using
a different key then an Autoincrement seems to help with scaling
Innodb on multiple CPU's (looking at this with a 4proc machine shows
that this issue show up at a later scaling point).

None of the other MySQL/3rd party engines behaved this way in
testing, so it looks to be a localized problem in Innodb. Blackhole
just gives you a base cost for an INSERT statement with no disk IO.

The test used 100K of rows and inserted them by splitting them across
the number of total connection (which gives you an estimate of
behavior by throwing a bigger machine/connections at a current problem).
Picture 1.png

  • Post a new comment


    Comments allowed for friends only

    Anonymous comments are disabled in this journal

    default userpic

    Your reply will be screened

    Your IP address will be recorded