On Sat, Jan 08, 2011 at 10:20:22PM -0600, Kevin Grittner wrote:
> One thing that would help a lot besides code review is performance
> testing. I did some months ago and I know Dan booked time on MIT
> benchmarking systems and got good numbers, but with the refactoring
> it would be good to redo that, and benchmarking properly can be very
> time consuming. Existing benchmark software might need to be tweaked
> to retry transactions which fail with SQLSTATE 40001, or at least
> continue on with out counting those in TPS figures, since
> applications using this feature will generally have frameworks which
> automatically do retries for that SQLSTATE.
I can certainly try to get a more complete set of DBT-2 results -- and
possibly even do that in a timely manner :-) -- but I doubt I'll have
time in the near future to do anything more comprehensive.
It would be great to have some more results beyond DBT-2/TPC-C.
Although it's certainly an interesting benchmark, it's known not to
exhibit any serialization anomalies under snapshot isolation. (And, of
course, it's seek-bound, so results may not be representative of
workloads that aren't.)
Dan
--
Dan R. K. Ports MIT CSAIL http://drkp.net/