Removing Legacy devices from a Guest in a HPVM environment

[Shawn Zehnder]

We currently have 2 Integrity Virtual Environments running HPVM 4.2 with 4 hosts and several guests. All hosts and guests are running HPUX 11.31. One environment contains an EVA4400 SAN and the other an EVA6400 SAN. Each SAN now contains 2 switches. We have some guests that were created using the legacy device files from the SAN presented LUNS. The LUNS have already been presented from the host as legacy devices which provides no multipathing. The guests have already been created so the presented disks are already in a VG. Is there any way to go back and modify the guests to use the agile devices since all multipathing is provided from the host. I found a command in the hpvmmodify MAN page but I'm not sure how this will effect the VG/LV layers on the guests. Please let me know if anyone has tried this before. Thanks for your help in advance.


Change the virtual disk with hardware address 0,0,4 to a different
physical device, /dev/rdisk/disk1:

# hpvmmodify -P myguest2x -m disk:scsi::0,0,4:disk:/dev/rdisk/disk1

[Torsten.]

>> as legacy devices which provides no multipathing.


This is not entirely true, because the documentation says that 11.31 does multipathing even with legacy device files.

I guess it should work if you remove the legacy file from the hpvm and add the agile device file.

[Shawn Zehnder]

Thanks for replying quickly. I'll assign points after I close this out. I saw that reference in the manual too. I was assuming that it was implying to use the agile device for the built-in multi-pathing. I'm going to eventually get another host set up for testing before trying anything. I currently only have one of the legacy devices presented for each lun. The manual does state that you can't use multi-pathing from the guests. It would need to be done from the host level.

Thanks

[Torsten.]

>> The manual does state that you can't use multi-pathing from the guests. It would need to be done from the host level.


This is true.


A document about the agile mass storage architecture (can't remember the document) says that multipathing is done even with legacy device files, but agile files are recommended.

[Ismail Azad]

Hi,

Just like to add one point as it was brought up. Not a VM expert but I can say one thing for sure. Legacy DSF's can take advantage of the mass storage stack of 11.31 for those who want to stick to strict backward compatibility. The parameter leg_mpath_enable when set to true can make this happen.

Regards
Ismail Azad

[Shawn Zehnder]

I have a call open with HP on this very subject. I would like to know how you would get the Legacy devices to be multi-pathed. I would prefer to use the agile if it's best practice. However, this would require the removal of the legacy devices from the guest before adding the agile devices. The funny thing is that I've searched the WEB and ITRC for the use of agile devices for this purpose. Most of the hits have suggested using lvm from the host to the guest. This seems very messy when it comes to Live Migration. It's too late to go that route. Not that I would want to.

[Ismail Azad]

Hi,

> . I would like to know how you would get the Legacy devices to be multi-pathed.

Not sure about a VM environment at all but since you posted the above line the following command should work on a server without virtualization.

scsimgr set_attr -D /dev/disk/disk3 -a leg_mpath_enable=true

Note:- This does not affect open Legacy DSFs.

Regards
Ismail Azad

[Cortes Albertino · Accepted solution]

Hello,

Regarding your initial HPVM question,

I had performed a small test a few months ago and it works, i.e.
removing the legacy backingstore device and then replacing it
with the agile device. I was able to delete the legacy (hpvmmodify -P guestname -d ....)
and then add (hpvmmodify -P guestname -a ...) the agile device instead without problems,
The impacted device was related to the vm guest boot disk and the guest was able to boot again.
In my case, the HPVM guest was also running HPUX 11.31 and AVIO
mode configured and i paid attention to keep the same "Bus Dev Ftn Tgt Lun" values.
( In my test, only vg00 was used/defined inside the guest virtual machine at time of test).

In the meantime it is preferable, from my point of view, to use the agile device
naming rather than legacy.

Also noticed page 96 in HPVM releases notes...

http://bizsupport1.austin.hp.com/bc/docs/support/SupportManual/c02545643/c02545643.pdf

Use of legacy device special files (DSFs) to define virtual storage will not be supported in a
future release.

Support for the use of legacy DSFs to define virtual storage (including virtual disks and
DVDs) will be dropped in a future release and customers should begin planning to use
persistent (agile) DSFs when defining their virtual storage devices. HP recommends the use
of persistent DSFs (for example, those with pathnames such as /dev/rdisk/disk##) for
better storage availability and reliability.
To check for the use of legacy DSFs, use the following command:
# hpvmstatus -P guestname -d | grep -w rdsk




Hope it helps,

Albertino

[Shawn Zehnder]

Thanks to everyone that replied. Cortes's reply sounds promising. I'll have a test environment set up soon for testing this out. I may leave this thread open until my testing has finished. Also, has anyone run performance testing on their luns tuning them with the scsimgr command?

Thanks, Shawn

[Rita C Workman]

The VMHost controls the devices.
So it doesn't matter from that side if you are using agile or legacy. The host should serve it up which path it needs to (agile or legacy).

Here's a great document that has a short section that explains this well...give it a read:

http://bizsupport1.austin.hp.com/bc/docs/support/SupportManual/c02018866/c02018866.pdf

Regards,
Rita