From: Michael Stone <mstone+postgres@mathom.us>
To: pgsql-performance@postgresql.org
Subject: Re: Inserts optimization?
Date: Fri, 14 Apr 2006 15:06:25 -0400
Message-ID: <20060414190620.GH32678@mathom.us> (raw)
In-Reply-To: <cone.1145037716.354055.283.5001@35st-server.simplicato.com>
References: <cone.1144888738.434948.92126.5001@35st-server.simplicato.com>
<443DA715.4060606@gmail.com>
<9549.1144903344@sss.pgh.pa.us>
<cone.1144954763.735087.95544.5001@35st-server.simplicato.com>
<20060414055923.GX49405@pervasive.com>
<cone.1145014225.784898.23930.1000@zoraida.natserv.net>
<20060414114911.GG32678@mathom.us>
<cone.1145037716.354055.283.5001@35st-server.simplicato.com>
On Fri, Apr 14, 2006 at 02:01:56PM -0400, Francisco Reyes wrote:
>Michael Stone writes:
>>I guess the first question is why 2 hot spares?
>
>Because we are using RAID 10
I still don't follow that. Why would the RAID level matter? IOW, are you
actually wanting 2 spares, or are you just stick with that because you
need a factor of two disks for your mirrors?
>>larger array with more spindles with outperform a smaller one with
>>fewer, regardless of RAID level (assuming a decent battery-backed
>>cache).
>
>Based on what I have read RAID 10 is supposed to be better with lots of
>random access.
Mmm, it's a bit more complicated than that. RAID 10 can be better if you
have lots of random writes (though a large RAID cache can mitigate
that). For small random reads the limiting factor is how fast you can
seek, and that number is based more on the number of disks than the RAID
level.
>>5 RAID5
>>2 RAID1
>>1 spare
>
>That is certainly something worth considering... Still I wonder if 2 more
>spindles will help enough to justify going to RAID 5. My understanding is
>that RAID10 has simpler computations requirements which is partly what
>makes it better for lots of random read/write.
If your RAID hardware notices a difference between the parity
calculations for RAID 5 and the mirroring of RAID 1 it's a fairly lousy
unit for 2006--those calculations are really trivial for modern
hardware. The reason that RAID 10 can give better random small block
write performance is that fewer disks need to be involved per write.
That's something that can be mitigated with a large cache to aggregate
the writes, but some controllers are much better than others in that
regard. This is really a case where you have to test with your
particular hardware & data, because the data access patterns are
critical in determining what kind of performance is required.
Mike Stone
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Reply to all the recipients using the --to and --cc options:
reply via email
To: pgsql-performance@postgresql.org
Cc: mstone+postgres@mathom.us
Subject: Re: Inserts optimization?
In-Reply-To: <20060414190620.GH32678@mathom.us>
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
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