From: | Peter Alban <peter(dot)alban2(at)gmail(dot)com> |
---|---|
To: | Kenneth Marshall <ktm(at)rice(dot)edu> |
Cc: | pgsql-performance(at)postgresql(dot)org |
Subject: | Re: Strange performance response for high load times |
Date: | 2009-06-18 19:42:47 |
Message-ID: | 477dfcc10906181242w4bb4de20sfe605f8de219ae92@mail.gmail.com |
Views: | Raw Message | Whole Thread | Download mbox | Resend email |
Thread: | |
Lists: | pgsql-performance |
So Ken ,
What do you reckon it should be ? What is the rule of thumb here ?
cheers,
Peter
On Thu, Jun 18, 2009 at 8:30 PM, Kenneth Marshall <ktm(at)rice(dot)edu> wrote:
> On Thu, Jun 18, 2009 at 08:27:02PM +0200, Peter Alban wrote:
> > Hi All,
> >
> > We are having a reasonably powerful machine for supporting about 20
> > databases but in total they're not more then 4GB in size.
> >
> > The machine is 2 processor 8 core and 8 Gig or ram so I would expect that
> PG
> > should cache the whole db into memory. Well actually it doesn't.
> >
> > What is more strange that a query that under zero load is running under
> > 100ms during high load times it can take up to 15 seconds !!
> > What on earth can make such difference ?
> >
> > here are the key config options that I set up :
> > # - Memory -
> >
> > shared_buffers = 170000 # min 16 or
> > max_connections*2, 8KB each
> > temp_buffers = 21000 # min 100, 8KB each
> > #max_prepared_transactions = 5 # can be 0 or more
> > # note: increasing max_prepared_transactions costs ~600 bytes of shared
> > memory
> > # per transaction slot, plus lock space (see max_locks_per_transaction).
> > work_mem = 1048576 # min 64, size in KB
> > maintenance_work_mem = 1048576 # min 1024, size in KB
>
> 1GB of work_mem is very high if you have more than a couple of
> queries that use it.
>
> Ken
>
> > #max_stack_depth = 2048 # min 100, size in KB
> >
> > # - Free Space Map -
> >
> > max_fsm_pages = 524298 # min max_fsm_relations*16, 6
> bytes
> > each
> > max_fsm_relations = 32768 # min 100, ~70 bytes each
> >
> > # - Kernel Resource Usage -
> >
> > max_files_per_process = 4000 # min 25
> > #preload_libraries = ''
> >
> > any ideas ?
> >
> > cheers,
> > Peter
>
From | Date | Subject | |
---|---|---|---|
Next Message | Kevin Grittner | 2009-06-18 19:54:38 | Re: performance with query |
Previous Message | Tom Lane | 2009-06-18 19:39:21 | Re: performance with query |