pg.ddx.io pgsql-performance@postgresql.org mailing list archive
help / color / mirror / Atom feedSSD endurance calculations
3+ messages / 3 participants
[nested] [flat]
* SSD endurance calculations
@ 2011-11-21 21:03 Christiaan Willemsen <cwillemsen@technocon.com>
0 siblings, 2 replies; 3+ messages in thread
From: Christiaan Willemsen @ 2011-11-21 21:03 UTC (permalink / raw)
To: pgsql-performance
Hey there,
We are looking at beefing up our servers with SSD's. Some of you did some interesting tests with the Intel 320. So the idea came to make a RAID10 with four 600GB models.
I did however do some calculations with the current database server (220GB database, expected to grow to 1TB by the end of next year). I specifically looked at /proc/diskstat at the read/write figures. From there I could see a read/write ratio of 3:1, and I also saw a wopping 170GB of writes per day (for a database that currently grows 1GB per dag). That seems like an insanely high figure to me! How come? We do mostly inserts, hardly any updates, virtually no deletes.
Secondly, I also looked at the reliability figures of the Intel 320. They show 5 years of 20GB per day, meaning that it will hold up for about 200 days in our system. RAID 10 wil make 400 days of that, but this seems hardly a lot.. Am I missing something here?
Kind regards,
Christiaan
^ permalink raw reply [nested|flat] 3+ messages in thread
* Re: SSD endurance calculations
@ 2011-11-22 16:59 Edgardo Portal <egportal2002@yahoo.com>
parent: Christiaan Willemsen <cwillemsen@technocon.com>
1 sibling, 0 replies; 3+ messages in thread
From: Edgardo Portal @ 2011-11-22 16:59 UTC (permalink / raw)
To: pgsql-performance
On 2011-11-21, Christiaan Willemsen <cwillemsen@technocon.com> wrote:
>We=
> are looking at beefing up our servers with SSD's. Some of you did so=
> me interesting tests with the Intel 320. So the idea came to make a RAID1=
> 0 with four 600GB models.</p><p> </p><p>I did however do some calcul=
> ations with the current database server (220GB database, expected to grow=
> to 1TB by the end of next year). I specifically looked at /proc/diskstat=
> at the read/write figures. From there I could see a read/write ratio of =
> 3:1, and I also saw a wopping 170GB of writes per day (for a database tha=
> t currently grows 1GB per dag). That seems like an insanely high figure t=
> o me! How come=3F We do mostly inserts, hardly any updates, virtually no =
> deletes.</p><p> </p><p>Secondly, I also looked at the reliability fi=
> gures of the Intel 320. They show 5 years of 20GB per day, meaning that i=
> t will hold up for about 200 days in our system. RAID 10 wil make 400 day=
> s of that, but this seems hardly a lot.. Am I missing something here=3F</=
> p><p> </p><p>Kind regards,</p><p> </p><p>Christiaan</p><div><p =
> style=3D"font-family: monospace; "> </p></div><p> </p><p> =
></p>=0A</body>=0A</html>
Is your WAL on a separate disk (or set of disks)?
Also, not sure you can fairly conclude that "RAID 10 will make 400 days
of that" -- I had read some posts here a few months back suggesting
that SSDs have been observed to fail very close to each
other in time in a RAID configuration.
^ permalink raw reply [nested|flat] 3+ messages in thread
* Re: SSD endurance calculations
@ 2011-11-23 15:19 Greg Smith <greg@2ndQuadrant.com>
parent: Christiaan Willemsen <cwillemsen@technocon.com>
1 sibling, 0 replies; 3+ messages in thread
From: Greg Smith @ 2011-11-23 15:19 UTC (permalink / raw)
To: pgsql-performance
On 11/21/2011 04:03 PM, Christiaan Willemsen wrote:
>
> Secondly, I also looked at the reliability figures of the Intel 320.
> They show 5 years of 20GB per day, meaning that it will hold up for
> about 200 days in our system. RAID 10 wil make 400 days of that, but
> this seems hardly a lot.. Am I missing something here?
>
The 320 series drives are not intended for things that write as heavily
as you do. If your database is growing fast enough that you're going to
hit a terabyte in a short period of time, just cross that right off the
list of possibilities. Intel's 710 series is the one aimed at your sort
of workload.
You can probably pull down the total write volume on your system by more
aggressively running VACUUM FREEZE shortly after new data is loaded.
PostgreSQL tends to write blocks even in INSERT-only tables several
times; forcing them to freeze early can eliminate several of them.
--
Greg Smith 2ndQuadrant US greg@2ndQuadrant.com Baltimore, MD
PostgreSQL Training, Services, and 24x7 Support www.2ndQuadrant.us
^ permalink raw reply [nested|flat] 3+ messages in thread
end of thread, other threads:[~2011-11-23 15:19 UTC | newest]
Thread overview: 3+ messages (download: mbox mbox.gz follow: Atom feed)
-- links below jump to the message on this page --
2011-11-21 21:03 SSD endurance calculations Christiaan Willemsen <cwillemsen@technocon.com>
2011-11-22 16:59 ` Edgardo Portal <egportal2002@yahoo.com>
2011-11-23 15:19 ` Greg Smith <greg@2ndQuadrant.com>
This inbox is served by DDX for PostgreSQL; see mirroring instructions
for how to clone and mirror all data and code used for this inbox