Re: Six PostgreSQL questions from a pokerplayer

On 7/5/09 11:13 PM, "Mark Kirkwood" <markir(at)paradise(dot)net(dot)nz> wrote:

> Craig Ringer wrote:
>> On Sat, 2009-07-04 at 11:51 -0700, Patvs wrote:
>>
>>
>>
>>> With 4 regular harddisks in RAID0 you get great read/write speeds, but the
>>> SSDs excel in IO/s and a 0.1ms access time.
>>>
>>
>> ... but are often really, really, really, really slow at writing. The
>> fancier ones are fast at writing but generally slow down over time.
>>
>>
>
> Also, (probably pointing out the obvious here) to be on the safe side
> you should avoid RAID0 for any data that is important to you - as it's
> pretty easy to get one bad disk straight from new!
>
> With respect to SSD's one option for a small sized database is 2xSSD in
> RAID1 - provided they are the *right* SSD that is, which at this point
> in time seems to be the Intel X25E. Note that I have not benchmarked
> this configuration, so no guarantees that it (or the Intel SSDs
> themselves) are as good as the various on-the-web tests indicate!

There is no reason to go RAID 1 with SSD's if this is an end-user box and
the data is recoverable. Unlike a hard drive, a decent SSD isn't expected
to go bad. I have deployed over 150 Intel X25-M's and they all work
flawlessly. Some had the 'slowdown' problem due to how they were written
to, but the recent firmware fixed that. At this point, I consider a single
high quality SSD as more fault tolerant than software raid-1.

Unless there are lots of writes going on (I'm guessing its mostly read,
given the description) a single X25-M will make the DB go very fast
regardless of random or sequential access.

If the system is CPU bound, then getting a SSD like that won't help as much.
But I'd be willing to bet that in a normal PC or workstation I/O is the
limiting factor. Some tuning of work_mem and shared_buffers might help
some too.

Use some monitoring tools (PerfMon 'Physical Disk' stats on windows) to see
if normal use is causing a lot of disk access. If so, and especially if its
mostly reads, an Intel X-25M will make a huge difference. If there is lots
of writes, an X-25E will do but its 40% the space for the same price.

>
> regards
>
> Mark
>
> --
> Sent via pgsql-performance mailing list (pgsql-performance(at)postgresql(dot)org)
> To make changes to your subscription:
> http://www.postgresql.org/mailpref/pgsql-performance
>