From: | Scott Marlowe <scott(dot)marlowe(at)gmail(dot)com> |
---|---|
To: | Stephen Frost <sfrost(at)snowman(dot)net> |
Cc: | Michael Kohl <michael(dot)kohl(at)tupalo(dot)com>, pgsql-performance(at)postgresql(dot)org |
Subject: | Re: High load, |
Date: | 2011-01-27 18:13:17 |
Message-ID: | AANLkTikTSkGKNXdqJOP_D78mXSRv-NXMZUnd4Q80qqwv@mail.gmail.com |
Views: | Raw Message | Whole Thread | Download mbox | Resend email |
Thread: | |
Lists: | pgsql-performance |
On Thu, Jan 27, 2011 at 10:54 AM, Stephen Frost <sfrost(at)snowman(dot)net> wrote:
> * Michael Kohl (michael(dot)kohl(at)tupalo(dot)com) wrote:
>> HDD: 2x 120 GB OCZ Vertex 2 SSD; RAID 1
>
> I'm amazed no one else has mentioned this yet, but you should look into
> splitting your data and your WALs. Obviously, having another set of
> SSDs to put your WALs on would be ideal.
Actually spinning media would be a better choice. A pair of fast
15krpm drives in a mirror will almost always outrun an SSD for
sequential write speed. Even meh-grade 7200RPM SATA drives will win.
> You should probably also be looking into adjustments to the background
> writer. It sounds like you're getting hit by large checkpoint i/o
> (if you turn on logging of that, as someone else suggested, you'll be
> able to corrollate the times), which can be helped by increasing the
> amount of writing done between checkpoints, so that the checkpoints
> aren't as big and painful. That can be done by making the background
> writer more aggressive.
This++. Increasing checkpoint segments can make a huge difference.
We run 64 segments in production and it's a world of difference from
the stock setting.
From | Date | Subject | |
---|---|---|---|
Next Message | Scott Marlowe | 2011-01-27 18:14:32 | Re: High load, |
Previous Message | Stephen Frost | 2011-01-27 17:54:22 | Re: High load, |