Bruce,
> > First, an interesting wierdness from the VACUUM FULL ANALYZE:
> > Analyzing elbs_clidesc
> > 2002-10-08 21:08:08 DEBUG: SIInsertDataEntry: table is 70% full,
> > signaling postmaster
> >
> > Huh?
>
> Well, you are dealing with elbs. That is the problem. ;-)
<grin> As you probably guessed, the purpose of these procedures is to
take a large amount (about 60mb) of not-normalized data from ELBS and
normalize it for our web-based case management system.
What's really frustrating about it is that we're only going to be doing
this for 2-3 months before we jettison ELBS for reasons that should be
obvious to you. But for those 2-3 months, the data transfer needs to
work well, and right now it doesn't even finish.
> You shoulnd't need that and it shouldn't lock up when it gets to 64.
> It
> should checkpoint and move on. The only problem with it being lower
> is
> that it will checkpoint more often.
Well, I'll try 128 and see if that helps any.
>
> > Rest of postgresql.conf params after my signature. All
> suggestions
> > are welcome. This server has been acting "sick" since I started
> with
> > it, under-performing my workstation and MS SQL Server. Either I've
> set
> > something wrong, or there's a hardware problem I need to track
> down.
> >
> > BTW, is there any problem for postgres in turning the fill access
> time
> > recorder in the host filesystem off? This is often good for a
> minor
> > performance gain.
>
> No problem.
>
> You might want to try pgbench and see if that works.
Yeah. I was planning on that -- as well as the postgresql.conf tuner
-- as soon as I can get through one data transfer so that I have a
little working time.
-Josh Berkus