On Fri, 07 Apr 2006 15:24:18 -0500
Scott Marlowe <smarlowe@g2switchworks.com> wrote:
> > See reply to Tom Lane :)
>
> I didn't see one go by yet... Could be sitting in the queue.
If it's not arrived by now - EXPLAIN ANALYZE doesn't tell me
anything :)
> Let us know if changing the fsync setting helps. Hopefully that's all
> the problem is.
fsync's already off - yes a bit scary, but our I/O is only about
500KB/sec writing.. the whole db fits in RAM / kernel disk cache, and
I'd rather have performance than security at this exact moment..
> Off on a tangent. If the aggregate memory bandwidth of the pSeries is
> no greater than you Xeon you might not see a big improvement if you
> were memory bound before. If you were CPU bound, you may or may not
> see an improvement.
I did look into the specs of the system, and the memory bw on the
pSeries was /much/ greater than the Xeon - it's one of the things that
really pushed me towards it in the end. I forget the figures, but it
was 3 or 4 times greater.
> Can you describe the disc subsystems in the two machines for us? What
> kind of read / write load you have? It could be the older box was
> running on IDE drives with fake fsync responses which would lie, be
> fast, but not reliable in case of a power outage.
Again, I'm confident that I/O's not the killer here.. the Xeon is a Dell
6850- hardware RAID1.. SCSI drives.
> > Multi-Opteron was the other thing we considered but decided to give
> > 'Big Iron' UNIX a whirl...
>
> It still might be a good choice, if it's a simple misconfiguration
> issue.
>
> But man, those new multiple core opterons can make some impressive
> machines for very little money.
So I see - we could buy two quad-opterons for the cost of renting this
pSeries for a month....
Cheers,
Gavin.
