Re: High end server and storage for a PostgreSQL OLTP system - Mailing list pgsql-performance

From Alex Turner
Subject Re: High end server and storage for a PostgreSQL OLTP system
Date
Msg-id 33c6269f05020105585b25794b@mail.gmail.com
Whole thread Raw
In response to Re: High end server and storage for a PostgreSQL OLTP system  ("Jim C. Nasby" <decibel@decibel.org>)
Responses Re: High end server and storage for a PostgreSQL OLTP system
List pgsql-performance
To be honest I've used compaq, dell and LSI SCSI RAID controllers and
got pretty pathetic benchmarks from all of them.  The best system I
have is the one I just built:

2xOpteron 242, Tyan S2885 MoBo, 4GB Ram, 14xSATA WD Raptor drives:
2xRaid 1, 1x4 disk Raid 10, 1x6 drive Raid 10.  2x3ware (now AMCC)
Escalade 9500S-8MI.

This system with fsync on has managed 2500 insert transactions/sec
(granted they are simple transactions, but still).

RAID 10 is a stripe of mirrors.  RAID 10 give you the best read and
write performance combined.  RAID 5 gives very bad write perfomance,
but good read performance.  With RAID 5 you can only loose a single
drive and rebuild times are slow.  RAID 10 can loose up to have the
array depending on which drives without loosing data.

I would be interested in starting a site listing RAID benchmarks under
linux.  If anyone is interested let me know.  I would be interested in
at least some bonnie++ benchmarks, and perhaps other if people would
like.

Alex Turner
NetEconomist


On Tue, 1 Feb 2005 05:27:27 -0600, Jim C. Nasby <decibel@decibel.org> wrote:
> On Tue, Feb 01, 2005 at 07:35:35AM +0100, Cosimo Streppone wrote:
> > >You might look at Opteron's, which theoretically have a higher data
> > >bandwidth. If you're doing anything data intensive, like a sort in
> > >memory, this could make a difference.
> >
> > Would Opteron systems need 64-bit postgresql (and os, gcc, ...)
> > build to have that advantage?
>
> Well, that would give you the most benefit, but the memory bandwidth is
> still greater than on a Xeon. There's really no issue with 64 bit if
> you're using open source software; it all compiles for 64 bits and
> you're good to go. http://stats.distributed.net runs on a dual opteron
> box running FreeBSD and I've had no issues.
>
> > >RAID10 will be faster than RAID1.
> >
> > Sorry Jim, by RAID10 you mean several raid1 arrays mounted on
> > different linux partitions? Or several raid1 arrays that
> > build up a raid0 array? In the latter case, who decides which
> > data goes in which raid1 array? Raid Adapter?
>
> You should take a look around online for a description of raid types.
>
> There's technically RAID0+1 and RAID1+0; one is a stripe of mirrored
> drives (a RAID 0 built out of RAID 1s), the other is a mirror of two
> RAID 0s. The former is much better; if you're lucky you can lose half
> your drives without any data loss (if each dead drive is part of a
> different mirror). Recovery is also faster.
>
> You'll almost certainly be much happier with hardware raid instead of
> software raid. stats.distributed.net runs a 3ware controller and SATA
> drives.
> --
> Jim C. Nasby, Database Consultant               decibel@decibel.org
> Give your computer some brain candy! www.distributed.net Team #1828
>
> Windows: "Where do you want to go today?"
> Linux: "Where do you want to go tomorrow?"
> FreeBSD: "Are you guys coming, or what?"
>
> ---------------------------(end of broadcast)---------------------------
> TIP 1: subscribe and unsubscribe commands go to majordomo@postgresql.org
>

pgsql-performance by date:

Previous
From: PFC
Date:
Subject: Re: Performance of count(*) on large tables vs SQL Server
Next
From: "Merlin Moncure"
Date:
Subject: Re: High end server and storage for a PostgreSQL OLTP system