Re: Postgres on SSD - Mailing list pgsql-general

From Greg Smith
Subject Re: Postgres on SSD
Date
Msg-id 4E458FB0.1050901@2ndQuadrant.com
Whole thread Raw
In response to Re: Postgres on SSD  (Vick Khera <vivek@khera.org>)
List pgsql-general
On 08/12/2011 04:24 PM, Vick Khera wrote:
> 2011/8/10 Ondrej Ivanič<ondrej.ivanic@gmail.com>:
>
>> Ups! Well spotted Tomas! The actual values are:
>> random_page_cost = 2
>> seq_page_cost = 1
>>
>>
> With the SSD I would set these to the same value of 1.  That's what I do.
>

That probably makes sense on your RAMSAN.  Sequential access on FusionIO
drives is at least 3X as fast as completely random though, and
worst-case it can be even slower relative to what a sequential scan can
deliver.  It's not the >50X difference seen on regular drives, but
there's an easily measurable gap.  I'm not sure if it's that the flash
cells deliver stuff faster when you read a sequential series from the
same cell of flash, or if it's just that there's less physical IOs
happening.

--
Greg Smith   2ndQuadrant US    greg@2ndQuadrant.com   Baltimore, MD
PostgreSQL Training, Services, and 24x7 Support  www.2ndQuadrant.us


pgsql-general by date:

Previous
From: Vincent Veyron
Date:
Subject: Re: Indicating DEFAULT values in INSERT statement
Next
From: Andy Colson
Date:
Subject: Re: Kudos