Subject: Re: Hardware/OS recommendations for large databases (
From: Scott Marlowe <smarlowe@g2switchworks.com>
To: David Boreham <david_list@boreham.org>
Cc: pgsql-performance@postgresql.org
In-Reply-To: <437B4773.1040404@boreham.org>
References: <3E37B936B592014B978C4415F90D662D01995F7B@MI8NYCMAIL06.Mi8.com>
	<33c6269f0511152115v1a34ab81j30f1f5df5a6c6956@mail.gmail.com>
	<437B3B18.1040509@commandprompt.com>  <437B4773.1040404@boreham.org>
Content-Type: text/plain
Content-Transfer-Encoding: 7bit
Message-Id: <1132160785.3582.60.camel@state.g2switchworks.com>
Mime-Version: 1.0
Date: Wed, 16 Nov 2005 11:06:25 -0600

On Wed, 2005-11-16 at 08:51, David Boreham wrote:
>  >Spend a fortune on dual core CPUs and then buy crappy disks...  I bet
>  >for most applications this system will be IO bound, and you will see a
>  >nice lot of drive failures in the first year of operation with
>  >consumer grade drives.
> 
> I guess I've never bought into the vendor story that there are
> two reliability grades. Why would they bother making two
> different kinds of bearing, motor etc ? Seems like it's more
> likely an excuse to justify higher prices. In my experience the
> expensive SCSI drives I own break frequently while the cheapo
> desktop drives just keep chunking along (modulo certain products
> that have a specific known reliability problem).
> 
> I'd expect that a larger number of hotter drives will give a less reliable
> system than a smaller number of cooler ones.

My experience has mirrored this.

Anyone remember back when HP made their SureStore drives?  We built 8
drive RAID arrays to ship to customer sites, pre-filled with data.  Not
a single one arrived fully operational.  The failure rate on those
drives was something like 60% in the first year, and HP quit making hard
drives because of it.

Those were SCSI Server class drives, supposedly built to last 5 years.

OTOH, I remember putting a pair of 60 Gig IDEs into a server that had
lots of ventilation and fans and such, and having no problems
whatsoever.

There was a big commercial EMC style array in the hosting center at the
same place that had something like a 16 wide by 16 tall array of IDE
drives for storing pdf / tiff stuff on it, and we had at least one
failure a month in it.  Of course, that's 256 drives, so you're gonna
have failures, and it was configured with a spare on every other row or
some such.  We just had a big box of hard drives and it was smart enough
to rebuild automagically when you put a new one in, so the maintenance
wasn't really that bad.  The performance was quite impressive too.