Subject: Re: SGI disk sorting algorithm, thoughts on disksort() lossage
To: Jonathan Stone <jonathan@DSG.Stanford.EDU>
From: Thor Lancelot Simon <tls@rek.tjls.com>
List: tech-perform
Date: 06/21/2002 17:44:45
On Fri, Jun 21, 2002 at 02:23:07PM -0700, Jonathan Stone wrote:
> 
> Obvious caveat: if this trick is used with a UBC-style system which is
> immature or poorly-tuned (or even de-tuned by the user), then we can
> still hit Manuel's scenario, The difference is that with the two-queue
> trick, the xterm running top would get almost exactly twice the
> page-in[ish] service rate it does now.

Actually, it's somewhat better than that, in practice.  You pull N requests
from each queue, not one request -- and a few clusters of page-ins can work
wonders.  Another method would take equal *amounts of I/O* from each queue;
that gets the pages in _very_ fast, if the other queue would clog things up
elsewise.

> 
> [... Sprite/NeXTt story: dont put more than one partition on a spindle]
> Which Sprite people?

My memory about that is quite vague, but I think it was John Hartman.  We
had a couple of late-night phone calls as I tried to replace Ultrix on
a large number of pmaxen with Sprite; building a big Sprite cluster that
was not just like the one at Berkeley was a more interesting exercise than
it might have seemed, and one of the better lessons that I took away from
it was to not run multiple filesystems per spindle if I could avoid it!

> My own workloads were happiest on machines with multiple filesystems
> (swap partition, plus one or more user filesystem) on each disk. We
> did carefully lay out the swap in the middle, tho, and the SCSI disks
> were fast for their day. Compared to RA-8x, anyway.

In fact, I do still often use a separate swap partition -- but these days,
I swap so little that it doesn't really matter much.  Even on a 2100 with
12MB of memory, the tradeoff in cycles/seeks for swap-through-FS versus
swap-to-partition came out the opposite way for me than it seems to have
for you -- though swap-to-partition with Sprite was not easy to achieve 
anyway.

> fwiw, I know IBM's logical volume-manager thingy for AIX made I/O
> really *glacially* slow if one ignored that advice.

Veritas doesn't like it much either; and neither does QFS.

-- 
 Thor Lancelot Simon	                                      tls@rek.tjls.com
   But as he knew no bad language, he had called him all the names of common
 objects that he could think of, and had screamed: "You lamp!  You towel!  You
 plate!" and so on.              --Sigmund Freud