Mark Townsend wrote:
> Serge Rielau wrote:
> Of course. There will always be shades of gray. But I can sleep at
night
> knowing that the TPC-C does _indeed_ run on a RAC cluster without the
> need for any special 'app infestation'. I also know that other
> non-benchmark applications run quite happily on RAC with the need for
> data locality. So in this case, if you are seeing something, I'm sure
> its there just to get maximum performance in a benchmark scenario, and
> is not necessarily indicative of a requirement in a real world
> scenarion, anymore than the faux partitioning in the DB2 benchmark is
> indicative of a requiremment in a real world scenario (FYI - is it the
> table limit thingy ?)
Yes it is. In DB2 + DPF there is no issue, which ironically makes teh
design simpler in TPC-C. On an SMP box the 4 byte row-id poses a scaleup
limit. UNION ALL. performance wise, gets very close to where I expect
partitioning to end up, which - no secret there - is in the works, and,
of course, preferable from a managability standpoint.
Cheers
Serge
--
Serge Rielau
DB2 SQL Compiler Development
IBM Toronto Lab
Received on Mon Jan 31 2005 - 05:56:27 CST
