Contrary to what others are saying, I don’t think that turning fsync off is killing you, you might get 25% more throughput at best.
The iostat output would indicate that you are getting just about all you can get out of it without getting faster drives or controllers.
Since it is mostly random read bound, more RAM in the system would certainly help reduce the load, as would an I/O subsystem with high random i/o throughput (quality flash drives, the best quality ones currently being Intel X-25 M until the next gen Micron, Samsung, Mtron, and SanDisk ones come out — do NOT use any that don’t have high random write perf).
I’m guessing that the cheapest thing to do is increase the RAM on the server — I think you can fit 16GB in an 1800, if not at least 8GB, and increase shared_buffers to between 10% and 25% of that. Upgrading to 8.3.x will further help.
On 1/8/09 6:41 AM, "Stefano Nichele" <stefano.nichele@xxxxxxxxx> wrote:
Glyn Astill wrote:
> --- On Thu, 8/1/09, Stefano Nichele <stefano.nichele@xxxxxxxxx> wrote:
>
>
>> From: Stefano Nichele <stefano.nichele@xxxxxxxxx>
>> Subject: Re: understanding postgres issues/bottlenecks
>> To: "Scott Marlowe" <scott.marlowe@xxxxxxxxx>
>> Cc: pgsql-performance@xxxxxxxxxxxxxx
>> Date: Thursday, 8 January, 2009, 8:36 AM
>> Find !
>>
>> Dell CERC SATA RAID 2 PCI SATA 6ch
>>
>> Running lspci -v:
>>
>> 03:09.0 RAID bus controller: Adaptec AAC-RAID (rev 01)
>> Subsystem: Dell CERC SATA RAID 2 PCI SATA 6ch
>> (DellCorsair)
>> Flags: bus master, 66MHz, slow devsel, latency 32,
>> IRQ 209
>> Memory at f8000000 (32-bit, prefetchable)
>> [size=64M]
>> Expansion ROM at fe800000 [disabled] [size=32K]
>> Capabilities: [80] Power Management version 2
>>
>>
>
> Hmm, the 64M / 6ch makes it sound like this card
>
> http://accessories.us.dell.com/sna/products/Controllers/productdetail.aspx?c=us&l=en&s=bsd&cs=04&sku=310-5975
>
Indeed I think it's this one. The server is a Power Edge 1800 Server.
> Which is a 6ch dell version of
>
> http://www.adaptec.com/en-US/support/raid/sata/AAR-2410SA/
>
> I have one on my smash pile. The only thing that makes me think otherwise is the 2 in "CERC SATA RAID 2" ...
>
>
>
>
>
>
--
Sent via pgsql-performance mailing list (pgsql-performance@xxxxxxxxxxxxxx)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-performance
