On Wed, 2005-04-13 at 00:00 +1000, Neil Conway wrote:
> Simon Riggs wrote:
> > During recent tuning of the TPC-C workload, I produced the following
> > patch to force COPY to leave some space in each data block when it loads
> > data into heap relations.
>
> I can't get too excited about incorporating changes designed solely to
> improve performance for the workload of a specific database benchmark.
Sometimes we tune for specific workloads, sometimes we need to tune to a
generic design pattern that effects many users. Unisys wanted to test
and tune a workload that would improve things for the most number of
users and I would say I support them in that.
Database benchmarks exist for two reasons:
- generate some great numbers
- they offer a generic workload that stresses PostgreSQL in pseudo-real
customer situations, but can be easily re-run, discussed, published and
dissected for real insight
The first reason helps people to accurately size systems and the second
reason helps them save them time and money....but without the first
people buy the wrong systems and waste money anyway. IMHO if you care
about the second, you should also care about the first.
Seriously, if you know a workload that better represents the majority of
performance critical database applications then I'll be happy to
consider tuning for that at another time. (Seriously).
> If the change has merit in some plausible "real world" situations, so be
> it -- but if not, I don't see the point.
Well, I've used PCTFREE many times with Oracle and DB2. Did it do any
good? Well, thats much harder, because you don't get as much chance to
tune specific issues like that in the real world, but yes, it makes a
difference for *some* real workloads. All? No.
Best Regards, Simon Riggs
