"I claim that "Linux swap to real DASD is good for one thing, and
that's to slow them down"  From your post I understand that is your
approach to tuning."

No, that's not true.

My approach is not to hurt the production systems, when a development,
test, or service machine gets in trouble.  In a shared environment you
can't throttle paging for a particular machine.  All I can do is limit
its maximum size (guest size plus vdisk).  The guest size is easy to
determine its impact.  Just bring the machine up, let it settle and
watch the paging.  Minor vdisk use, also isn't a problem.  But when
something goes off, and uses/abuses all the vdisk it can get, VM starts
paging and all machines are impacted.

Perhaps, this is a good reason to have separate LPARs.  One for
production Linux and the other for all the other Linux stuff.  Being a
VM bigot, that was a hard thing for me to say. <G>

Try this little task:

1.  Logon to MAINT
2.  Def stor (to the size of your LPAR)
3.  Run the following:

/* use up storage */
a = 'a'
do forever
   a = a || 'a'
end

See if any users complain about poor response time.  If you have
sufficient "extra" resources to handle it, great.  But there will be a
time when it would be a big impact.  (really, you shouldn't run it)

I guess part of the vdisk abuse problem, is how large each shop makes
the vdisk.

I use (unless otherwise needed)  16MB/ 32MB/ 64MB.  That is 112 MB
total, if being abused.  Now that I'm in a 3 GB LPAR, I doubt it would
ever be a problem.  But when I was on a 1 GB LPAR that was actively
paging and my vdisk (note, only one) was 250MB (after all, it is free).


Of course, in the VM world, we had this problem ever since vdisk was
introduced.  Application programmers in a 16 MB CMS machine, grabbing
512 MB vdisk and abusing them.  My little 256MB IBM 9121 was on its
knees.  And without a performance monitor, finding that it was a vdisk
abuse problem was rather difficult.



Tom Duerbusch
THD Consulting

>>> [EMAIL PROTECTED] 2/23/2007 3:30 PM >>>
On 2/23/07, Tom Duerbusch <[EMAIL PROTECTED]> wrote:

> In a memory constrained system, it may be better (definitely for
> test/development systems) to have swap go to dasd.  Then, you are
only
> impacting that machine.  When you have a big impact on the paging
> system, almost everyone (including production) can be impacted.

I claim that "Linux swap to real DASD is good for one thing, and
that's to slow them down"  From your post I understand that is your
approach to tuning. The sad part is that you slow them down all the
time when they swap, also when enough resources available. To avoid
that, you need to make the servers large enough that they normally
don't swap... and so you become more memory constrained. <img
src=death-spiral.gif>

> Actually, Mark, we agree on the basics of vdisk/swap.  But I have
never
> seen a paging system that couldn't be swamped.  How many times have
I
> PEEKed a file, should have been 10,000 lines, but it was 1,000,000
lines
> (or 10 million lines), and watch the paging system, deal with my
error.

If you really measured that your paging subsystem was holding you back
there, then you should probably improve that. Properly configured,
z/VM can page pretty heavy without hurting itself. Very likely that it
was the spooling subsystem that held you back because we sip slowly on
spool files.

Rob
--
Rob van der Heij
Velocity Software, Inc
http://velocitysoftware.com/

----------------------------------------------------------------------
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