I use to be in the group that had vdisk and real disk for swap.  It just
seemed to make sense.  Part of that sense was that I had some old RAMACs
that I could use for DASD swap.  But no more.  I only have one class of
dasd, DS6800 Ficon attached.

It became apparent, that if I was going to dedicated disk space for
swap areas, that is is better to give that disk space to the paging
system, and just add another VDISK or two (each at different and lower
priorities).  Worse case is I use exactly the same amount of disk space
either way.  Best case, I gain.

The only negative was that adding/'removing a Linux swap volume is much
easier than adding/removing a CP paging device.

But back on the initial topic.  I assume that vdisk isn't an option (or
not running under VM).  Then the swap volumes really shouldn't be on the
same FCP path (perhaps can't help that unless you have multiple SANs),
or on the same controller, same raid arrray as your Linux image.
Sometimes it can't be helped.  But if Linux is starting to eat up swap
space, then what is using the virtual memory and how it it being loaded?
 Most likely, it is reading from somewhere.  Perhaps a large database
query.  From a performance perspective, I wouldn't want my, now active
swap disks, to be on the same path as the data that is being read that
is causing the swapping.  I'm not talking about Linux caching, but some
application that is reading data and storing it in virtual storage.

You also have to consider if this occurs, shouldn't the process be
terminated or Linux cycled?  The plan for a runaway task is to terminate
the task, not give it infinite resources.

And then there are valid reasons.  I have some nice, tight, Linux
machines, that went bonkers when I tried to apply a service pack.  The
real solution was to change the virtual storage size from 48MB to 256MB
and then apply service.  Then change back to 48 MB and IPL again.  But
that might not be an option in all cases.

Tom Duerbusch
THD Consulting

>>> [EMAIL PROTECTED] 1/25/2007 2:20 PM >>>
On Jan 25, 2007, at 1:26 PM, Rob van der Heij wrote:

> On 1/25/07, Terry Spaulding <[EMAIL PROTECTED]> wrote:
>
>> I am looking for a best practice when you need a swap file/
>> partition and
>> would like to hear from all on what they are doing.
>
> The best swap disk for Linux on z/VM is VDISK, so I don't think this
> is an issue.
> People who use real disk for Linux swap only do so to slow down
their
> server and probably do not care for slow or even more slow. Or they
> are very certain they will never swap, and in that case I/O
contention
> for the device is no issue either.

I'd say this is not quite true.  Place as much swap as you usually
use on VDISK.  Place some more at lower priority on VDISK if you can
spare the VDISK.  If you're out of VDISK/are experienceing real
storage constraints, then think about a third tier on actual DASD,
particularly if you have ill-behaved applications that usually use X
amount of memory but sometimes for no obvious reason suddenly want 4X
for a while.  The question you have to ask is whether poor
performance is better than application failure.  The answer isn't
always "yes", particularly if the failure is because the app in
question leaks memory so eventually you'd have to restart it anyway.

Adam

----------------------------------------------------------------------
For LINUX-390 subscribe / signoff / archive access instructions,
send email to [EMAIL PROTECTED] with the message: INFO LINUX-390
or visit
http://www.marist.edu/htbin/wlvindex?LINUX-390

----------------------------------------------------------------------
For LINUX-390 subscribe / signoff / archive access instructions,
send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit
http://www.marist.edu/htbin/wlvindex?LINUX-390

Reply via email to