Re: One particular large database application - Mailing list pgsql-general

From Curt Sampson
Subject Re: One particular large database application
Date
Msg-id Pine.NEB.4.43.0204231327310.447-100000@angelic.cynic.net
Whole thread Raw
In response to Re: One particular large database application  (Michael Loftis <mloftis@wgops.com>)
List pgsql-general
On Mon, 22 Apr 2002, Michael Loftis wrote:

> You're still limited to onlya few controllers in a system.

Well, I usually can spare two or three PCI slots, so that's six or
eight controllers (and thus drives), at any rate. If I really needed
more drives or storage capacity on a system at that point, it's
probably best to move to an external disk array anyway.

> There's also
> the issue of concurrency.  IDE drives handle one, and only one request
> at a time.  A SCSI drive can (and usually is) be issued commands with
> tags allowing more than one active/pending command at a time.

Sure. But still, send four requests to a SCSI drive, and four
requests to four IDE drives, and see which comes back faster.

> The better drives
> all have 'smart' cache controllers on them that re-order the pending
> ocmmands in a way that optimizes response and throughput in an
> intelligent manner (I think seagate is calling it serpentine seek, but
> whatever it is called..)

Many operating systems do this, too, though obviously they may not
be able to do it quite as well as the controller can.

> SCSI is more expensive, there's no doubt, but for larger environments,
> there are clear benefits.

Indeed. I've never argued against this.

I am well aware of the various advantages of SCSI drives. I'm also
aware of what they cost. It's just that there's a particular person
on the list who seems to think a less cost-effective solution would
somehow be better.

> Also in a SCSI system, a failed drives
> electronics will, in most cases, isolate itself.  In an IDE system a
> failed drive will, at the least, make the other drives on that chain
> unavailable, and in many cases make any other drives on the same
> controller unavailable.

Not a big deal; nobody interested in performance is going to put
more than a single drive on an IDE controller anyway.

cjs
--
Curt Sampson  <cjs@cynic.net>   +81 90 7737 2974   http://www.netbsd.org
    Don't you know, in this new Dark Age, we're all light.  --XTC


pgsql-general by date:

Previous
From: Michael Loftis
Date:
Subject: Re: One particular large database application
Next
From: Lincoln Yeoh
Date:
Subject: Re: One particular large database application