Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!swrinde!mips!dimacs.rutgers.edu!rutgers!cbmvax!grr
From: grr@cbmvax.commodore.com (George Robbins)
Newsgroups: comp.unix.ultrix
Subject: Re: A question about swap
Keywords: swap
Message-ID: <22486@cbmvax.commodore.com>
Date: 15 Jun 91 21:46:43 GMT
References: <1991Jun14.184609.21178@mlb.semi.harris.com>
Reply-To: grr@cbmvax.commodore.com (George Robbins)
Organization: Commodore, West Chester, PA
Lines: 39

In article <1991Jun14.184609.21178@mlb.semi.harris.com> dcb@dave.mis.semi.harris.com writes:
> On one of our 5500s, we have configured over 500M of swap space. Someone
> recommended using the 'a' and 'b' partitions over five drives, with the
> a/b partitions combined into a larger 'a' partition.
> 
> Does this buy us much over a single 500M swap partition? Is the OS smart
> enough to  distribute the load over several spindles?

Supposedly the Berkeley code does distribute swap usage over the multiple
drives in a way that improves performance.

On the to note is that swap space is kind of a dinosaur from the days of
7*'s and 10 million students.  If you have enough memory, swap usage is
minimal and even paging low (unless you're running X).

Most the requirement for large swaps spaces stems from the problem that
Ultrix allocates swap space before it actually needs it.  The other is
megalithic software that depends on virtual memory to fit at all.  Some
of this is well behaved, some will cause a paging frenzy.

I'd suggest you use pstat/vmstat/monitor to evaluate how many paging/
swapping event occur per second.  If the number is less than 1, it's
a don't care.  If the number is a little larger you need to contemplate
whether you will get better action with a single "swap drive" with
only self-contention or spreading it over several shared drives.  If
the number is large, see if you can afford mega-memory.

Large here should be considered relative to the number of random
read/write requests the controller/drive combination can perform
per second, less whatever useful disk I/O you need to get done.

You might also want to look at that Ultrix tuning paper that was posted
here not long ago, it appears that Ultrix may be surprisingly poorly
tuned for today's RISC systems...

-- 
George Robbins - now working for,     uucp:   {uunet|pyramid|rutgers}!cbmvax!grr
but no way officially representing:   domain: grr@cbmvax.commodore.com
Commodore, Engineering Department     phone:  215-431-9349 (only by moonlite)