VIOS - dedicated vhost for rootvg disks?

[unprophete]

For disk i/o performance on client LPAR does it matter rootvg and datavg disks are mapped on the same or on separates vhost devices on VIOSes?

What IBM recommends in such case?

 

[juredd1]

I don't have any technical details or any links providing info but from personal experience with IBM Support when I was setting up my VIOServer I was creating on vhost per physical disk (Yea a bit over kill I know), and I was instructed by IBM support to use one vhosts for all my disks/VTD devices related to one client LPAR. I am thinking there is a limit of how many disks/VTD devices per vhosts but remember the magic number.

Hope that helps.

 

[p5wizard]

With a queue depth value at the default of 3, and the number of queue elements per vhost/vscsi pair being 512, you could have close to 100 virtual disks on one vhost/vscsi pair.

But queue depth should probably be tuned to a bigger value so you should use a lower maximum especially if the backing devices are SAN.

There is some info about it in this redbook (which I googled for you ;-)):

http://www.redbooks.ibm.com/abstracts/sg247940.html?Open

(see page 81)



HTH,

p5wizard

 

[p5wizard]

That being said, I'd still use a dedicated vhost/vscsi pair for rootvg virtual disks and another for app/data VG disks.

HTH,

p5wizard

 

[unprophete]

In my case both rootvg and datavg come from the same SAN storages (in total 10 LUNs - 5 from one storage and 5 from the other). The queue_depth for each hdisk is setup to 10. They queue_depth is configured on hdisks on VIOS side and on Client LPAR as well. Is that correct?

Does one vhost for all 10 LUNs sounds still ok in my case?

Now, I will also look into the redbook P5wizard attached.

 

[p5wizard]

Yes, queue depth should be the same on client side and on VIOS side.

In the redbook, the suggested maximum for SAN LUNs with queue depth of 10 is 26 devices per vhost/vscsi. So I'd assume you are OK.

HTH,

p5wizard