Pat - sure, any intelligent code paths will help.  Certainly in a
virtualized environment including but not limited to system z resources
are being shared intensely.  The q4 problem (maybe I should trademark
it!) -- errant q3 -- is insidious and damaging.  

These aren't grandpa's CMS machines with small working set sizes.  The
machines which are waking up needlessly due to application layer code
typically have very large WSSes. So regardless of path length you have
these virtual beasts competing against each other and legit work
inducing unneeded paging, storage management, etc.
And what is most expensive dollar resource? Unless you are getting the
deal of the millennium it's system Z memory, not the IFLs.

In general I have found you cannot tune your way out this with SRM
values or other CP settings.  Keeping your Linux virtual machine size as
small as you can while providing decent service is advisable, but it
doesn't keep them out of q4. A large DASD paging farm and appropriate
xstore values helps contain this but it not a fix.

I fail to see why applications are reluctant to determine what
environment they are in and make decisions accordingly. I know and
understand the rationale behind agnostic code but the entire system z
solution for Linux is being hurt by this.

It just seems unreasonable for any IBM product to be insensitive to
running in a virtual machine.  The kernel certainly knows, hey, it even
announces it at boot time!
David Kreuter


-------- Original Message --------
Subject: Re: mono keep guest active
From: Patrick Spinler <[email protected]>
Date: Tue, August 17, 2010 5:51 pm
To: [email protected]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

David Kreuter wrote:
> The non-hostile list is quite short unfortunately. For the most part
> Oracle is not hostile and queue drops nicely.
> Getting vendors including IBM to:
> 1. acknowledge the problem is hard.
>
> 2. once acknowledged repairing (woops, I mean adding a feature) doesn't
> happen quickly or for that matter often.
>
> In my view it is not criminal or heretic for code to acknowledge its
> virtual surroundings. But lots of apps people think otherwise.
>
> People we just want all our virtual machine children to play and share
> nicely. Give up when you do not have actual work, you will get your turn
> when needed, really you will. Is that too much to ask?
> David Kreuter
>

It seems to me that this issue has certain parallels to the current and
long running debate about linux kernel power management hacks targeting
embedded devices (e.g. android wake locks)

Specifically -- applications are very frequently crappy, and fixing them
all, or even a significant fraction of them, is significantly unlikely.
 Ergo, what, if anything, could a linux kernel do to reign in
misbehaving apps?

Android's answer is to sleep regardless of what the apps say, with a
privilege limited mechanism that blocks sleeping. Privs are only
granted to apps the admin (or android packager) deems truely critical
like the radio / phone apps.

Would some similar sort of mechanism help for virtualization?
(complete, uninformed speculation here) Perhaps a kernel mechanism to
limit wakeups in the case that no cpu is seen to be consumed, or the
like?

- -- Pat

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (Darwin)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/

iEYEARECAAYFAkxrBHsACgkQNObCqA8uBswOzwCeN8Sdm59uWxiJXRJiYT60FYX7
4h8AnixYLgrj2+uGx4O2DgD4yI9ornI+
=0pfK
-----END PGP SIGNATURE-----

----------------------------------------------------------------------
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 more information on Linux on System z, visit
http://wiki.linuxvm.org/

----------------------------------------------------------------------
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 more information on Linux on System z, visit
http://wiki.linuxvm.org/

Reply via email to