>>> On Tue, Jan 15, 2008 at 11:05 AM, in message
<1200416720.4742.6.camel@arc-dknightlx.wsicorp.com>, Doug Knight
<dknight@wsi.com> wrote:
> We tried reducing the memory footprint of the postgres processes, via
> shared_buffers (from 30000 on Linux to 3000 on Windows),
I would never go below 10000. 20000 to 30000 is a good start.
> max_fsm_pages (from 2000250 on Linux to 100000 on Windows)
> max_fsm_relations (from 20000 on Linux to 5000 on Windows)
Figure out what you need and use that. Low values can cause bloat.
Check the output of VACUUM ANALYZE VERBOSE (for the database)
for a line like this:
INFO: free space map contains 717364 pages in 596 relations
> and max_connections (from 222 on Linux to 100 on Windows).
If you don't need more than 100, good.
> Another variable we played with was effective_cache_size
> (174000 on Linux, 43700 on Windows).
Figure out how much space is available to cache data,
counting both the shared buffers and the Windows cache.
This setting has no affect on memory usage, just the planner.
> On Mon, 2008-01-07 at 19:49 +0500, Usama Dar wrote:
>
>> Doug Knight wrote:
>> > We are running the binary distribution, version 8.2.5-1, installed on
>> > Windows XP Pro 32 bit with SP2. We typically run postgres on linux,
>> > but have a need to run it under windows as well. Our typical admin
>> > tuning for postgresql.conf doesn't seem to be as applicable for windows.
>>
>> So what have you tuned so far? what are your current postgresql settings
>> that you have modified? What are your system specs for Hardware, RAM ,
>> CPU etc?
I would add that you should post what problems you're seeing.
>> Is there a place where I can find information about tuning
>> postgresql running on a Windows XP Pro 32 bit system? I
>> installed using the binary installer. I am seeing a high page
>> fault delta and total page faults for one of the postgresql
>> processes.
Are those "hard" or "soft" page faults. There's a big difference.
Review Windows docs for descriptions and how to check.
You may not be able to do much about soft page faults,
and they may not have much impact on your performance.
We abandoned Windows for database servers after testing
identical processing on identical hardware -- Linux was twice
as fast. If you really need to run under Windows, you may need
to adjust your performance expectations.
-Kevin
