Without PAV, having Linux issue 3 I/Os to the same physical
volume will NOT help performance. If one of those I/O must
be satisified by activity to a real disk, all of the remaining I/O
queued on the device will wait, even those with data residing
in cache who could be satisfied immediately.

And the fact that VM "orders the I/O by seek address" really
makes no difference with today's DASD. It's mostly a waste
of cycles.

The inability for VM (and linux guests under VM) to efficiently
use large disks is the issue with VM not having PAV -
installations with SHARK or Symetrix are better off defining
many smaller "PHYSICAL" devices.  This is how VM gets around not
having PAV.  Thus when linux issues 3 I/O, they are to 3
different "physical" devices, VM can drive all 3 I/O, and those
I/O directly satisfied by the cache do not have to wait for
those I/O needing disk access.

In addition, and specifically for databases where
high bandwidth is needed, not only do you want the LVM'ed
volumes to be on different physical volumes, but also having
different channel paths, and different internal paths on the
storage controller when possible.

>
>I always thought, and I may be wrong, that one of the main
>advantages to having either a PAV or individual minidisks was
>so that instead of having 1 queue for the device, you now have
>many.  If your first I/O in the queue needs to do a physical
>I/O, then all the other I/Os wait.  If most of the I/Os have
>their data already in cache, then those I/Os will be done at the
>same time as the physical I/O.  I know we get about 95% of all
>our I/Os on MVS from cache.
>
>Eric Bielefeld
>Sr. MVS Systems Programmer
>P&H Mining Equipment
>Milwaukee, WI
>414-671-7849
>[EMAIL PROTECTED]
>
>>>> [EMAIL PROTECTED] 03/26/03 02:33PM >>>
>Physically, that's true -- ultimately there is only one physical
>I/O in progress.  However, by splitting up the disks into a
>larger number of small chunks and presenting them to Linux in
>the virtual machine configuration, the Linux system sees the
>smaller minidisks as separate volumes, and thus schedules
>multiple I/Os to what it thinks is multple devices.  CP
>coordinates all the actual disk I/O and everybody wins.
>
>If you can, stripe the pieces across multiple physical voluems,
>but it's not as important.
>
>-- db
>
>David Boyes
>Sine Nomine Associates







"If you can't measure it, I'm Just NOT interested!"(tm)

/************************************************************/
Barton Robinson - CBW     Internet: [EMAIL PROTECTED]
Velocity Software, Inc    Mailing Address:
 196-D Castro Street       P.O. Box 390640
 Mountain View, CA 94041   Mountain View, CA 94039-0640

VM Performance Hotline:   650-964-8867
Fax: 650-964-9012         Web Page:  WWW.VELOCITY-SOFTWARE.COM
/************************************************************/

Reply via email to