"Euler Taveira de Oliveira" <euler@timbira.com> writes:
>> I recall Heikki was doing lots of tests with DBT2 for EnterpriseDB
>> comparing those two releases like that; anybody know of a good summary I
>> could utilize there? I thought that was one of the benchmarks that got a
>> 20-30% speedup on going to 8.3 because it really took advantage of HOT in
>> particular.
If you're doing large TPC-C runs then the phantom-command-id and packed
varlena changes give you about 9% space savings which translates surprisingly
nicely into about 9% TPM increase. (This is for a specific schema, if you
dumbify your schema with char(1)s and numerics you could see a bigger
difference)
HOT has a *huge* effect on how long your can run the benchmark before
performance starts to droop. However the minimum run time for TPCC is only 2
hours and for large runs that's not enough for vacuum-related issues to kick
in.
Also, smoothed checkpoints have a *huge* effect but TPC-C is based on 95th
percentile response times and the checkpoints only affect about 1% of the
transaction response times.
I think TPC-E will make both of these major improvements much more important.
I suspect it would be hard to get 8.2 to even pass TPC-E due to the checkpoint
dropouts.
--
Gregory Stark
EnterpriseDB http://www.enterprisedb.com
Ask me about EnterpriseDB's On-Demand Production Tuning