Re: System V IPC on Windows

Lee Keel wrote:
>> -----Original Message-----
>> From: pgsql-general-owner(at)postgresql(dot)org [mailto:pgsql-general-
>> owner(at)postgresql(dot)org] On Behalf Of Magnus Hagander
>> Sent: Thursday, November 08, 2007 1:30 PM
>> To: Kevin Neufeld
>> Cc: Tom Lane; pgsql-general(at)postgresql(dot)org
>> Subject: Re: [GENERAL] System V IPC on Windows
>>
>> Pg on win32 is 32-bit. It has a total address space of 2Gb, minus the OS
>> overhead, minus the code, minus local memory etc. You're just not going
>> to fit that much in the address space.
>>
>> There are fixes to reduce the memory usage of the postmaster (which is
>> likely what runs out first, unless you have a large work_mem) in 8.3,
>> but you're still right up against the wall with that large
>> shared_buffers. Plus, as I said in my other email, you're likely not
>> seeing any performance gain from such a large shared_buffers anyway. If
>> you are, you're seeing something new, and we definitely need to find out
>> why.
>>
>>
>> //Magnus
>>
>>
> [Lee Keel]
>
> I can't do any benchmarks because I keep getting errors. But I have dropped
> this value down and I am not getting the out of memory errors any more. I
> was trying to solve other problems by bumping this way up but it seemed to

You could try benchmarking in-between levels, like comparing 128Mb to
512Mb. A couple of those should show you a trend.

> just cause more problems. I found several things in documentation that says
> that bumping the shared_buffers\work_mem up to over 1gb was fine. Is that
> strictly for linux boxes?

No, it's valid for most platforms that aren't Windows. But it's
generally not valid for Windows.

//Magnus