Re: Paging file

["Steve Landess" <steve_landess@xxxxxxxxxxx>]

>----- Original Message ------
>From: "James Rich" <james@eaerich.com>
>To: <midrange-l@midrange.com>
>Sent: Tuesday, October 29, 2002 12:49 PM
>Subject: Re: Paging file
> I believe (i.e. I don't know) that in this case a dedicated swap partition
> should actually outperform the "spread out over all the disks" case
> because the swap data is "compact", meaning that it is contiguous.  One
> disk arm reading a continuous block of data should beat many arms seeking
> all over several disks.
> But without benchmarks, tests, case studies for specific applications,
> etc. we will probably never know.
> Of course arguments to the contrary are welcome and read with interest.

James, I disagree (and I am contrary <grin>)...

In fact, I have seen cases where, because of a concentration of data for a
given object (usually a data file) on a particular unit, you could end up
having serious disk-arm contention problems on one or more units.  This
was/is a common occurrence when new units were added to an existing system
without save/reload.  So, unless your dedicated swap area is scattered
across all units (or a signficant number of dedicated units), you could
possibly run into this situation.

Back before the STRASPBAL command, the only way that I knew how to truly
solve this data concentration problem was to save the entire system, scratch
& reformat, then reload EVERYTHING (what IBM called a "scatter-load").  Of
course, there were always kludgey ways to move data around and get it more
balanced, but to me they weren't very scientific.

I don't think you don't want to have to wait on queued-up requests for data
where a small number of disk units is handling a disproportionate share of
the workload.  And if you have 1,000 users sharing a dedicated swap space,
this could be a reality.

JMHO,
Steve Landess
Austin, Texas
(512) 423-0935

Quitting smoking is the easiest thing I have ever done.  I ought to know,
because I have done it at least 1000 times...(Mark Twain)