Online VMs Migration

[David Dilly]

Hello,

SLVM is supported for online migration of VMs (Without Service Guard) ?

Thanks.

[Dave Olker]

Hi David,

I don't think you can do this without Serviceguard since SG is responsible for ensuring only one node is using the SLVM volumes at a time.

Looking at the HPVM 4.2 Installation and Configuration guide (page 201):

Integrity VM now supports the migration of virtual machines that are using LVM logical volume (lvol) backing stores configured with avio_stor, also known as shared LVM (SLVM). To enable SLVM functionality, Serviceguard A.11.19 and patch PHSS_40152 must be installed on the VM Host systems. In addition, for virtual machines using avio_stor adapters, version B.11.31.0910.01 of the HostAVIOStor software bundle must be installed. This software is available for download from http://software.hp.com/.

Hope this helps,

Dave

[P Muralidhar Kini]

Hi David,

Integrity VM now supports the migration of virtual machines that are using
LVM logical volume (lvol) backing stores configured with avio_stor,
also known as shared LVM (SLVM).

Check the following link for more details -
http://docs.hp.com/en/T2767-90797/ch09s01.html

Regards,
Murali

[David Dilly]

I asked this question because I'd like to replicate data between two EVA storage arrays.

EVA-1 is a bay in a site A and the other (EVA-2) is in a site B.

On each site there will be a VM Host.

The goal is to configure a VM Guest with secure data with LVM mirrored between the 2 sites.

The LVM mirror will be provided by the VM Hosts.

The two VM Hosts have access to VG that contains these LV Mirror mode thanks to "Shared".

The purpose of this configuration is that in the event of a loss of site A, I can start my VM guest on another site.

But I would like also that the online migration works to.

Does this configuration is supported:

SLVM across two VM Host on two different EVA storage array with online migration of VMs Guest.

Thank you.

[Eric SAUBIGNAC]

Bonsoir David,


I understand that :

- you need to do LVM Mirror to protect LVM data of the VMs

- LVM Mirror is not supported in a VM

- so to protect LVM data with LVM mirror you need to configure it at host level

==> Thats' why you plan to use SLVM to do online migration. Right ?


Now suppose that "LVM Mirror" is now supported in a VM ???

Check the beginning of page 90, in the HPVM 4.2 Release Notes ( http://h20000.www2.hp.com/bizsupport/TechSupport/CoreRedirect.jsp?redirectReason=DocIndexPDF&prodSeriesId=4146186&targetPage=http%3A%2F%2Fbizsupport2.austin.hp.com%2Fbc%2Fdocs%2Fsupport%2FSupportManual%2Fc02123997%2Fc02123997.pdf ). We can read :

----------------------------------
NOTE: LVM mirroring in a VM is supported with 0909 or later versions of HostAVIOStor (on the VM Host) and GuestAVIOStor (on the guest).
----------------------------------


If LVM mirror is really supported in a VM Guest, you will no more need to use SLVM to do VM online migration in the situation you have described. Just move LVs from the host inside the VM.

In fact I am a little bit surprised by this release note and I have asked HP to confirm.

Eric

[Dave Olker]

> If LVM mirror is really supported in a VM
> Guest, you will no more need to use SLVM
> to do VM online migration in the situation
> you have described. Just move LVs from the
> host inside the VM.
>
> In fact I am a little bit surprised by
> this release note and I have asked HP to
> confirm.

Yes, LVM mirroring is supported inside the VM guest. This is typically used when a VM guest is acting as a Serviceguard Node in a cluster where other standalone (i.e. physical) servers are the other cluster members and there is a need to do LVM mirroring as part of a package.

If the only goal is to have the LVM data mirrored, the recommendation remains to do the LVM mirroring on the VM host and present the mirrored LV to the guest. That way the host handles all the mirroring tasks and the guest simply sees an LV.

Keep in mind, if you're going to do LVM mirroring from inside the HPVM Guest then the HPVM Host needs to present the whole disks to the HPVM Guest, not the mirrored LV. That way the HPVM guest sees the entire disk and will handle the LVM mirroring tasks inside the guest just as a standalone physical server would.

Regards,

Dave

[Eric SAUBIGNAC · Accepted solution]

"the recommendation remains to do the LVM mirroring on the VM host "

The question behind this recommendation is : why. After all is it supported or not ?

- Performance reasons : datas have to go through the virtualization layers twice ?

- Virtual IO, whatever emulated scsi or avio, do not handle correctly hardware events ?

- Other ?


Facing this release note and the recommendation you intelligently point (and that lies between documentation lines) I feel like I was at the very beginning of HP virtualization : HP's speech is not clear enough with this subject. But I guess it is too much to ask HP to point their own weakness. I can understand this in a competitive world, but in fact, as a technical engineer... I refuse to.



Doing "LVM mirroring from inside the HPVM Guest then the HPVM Host needs to present the whole disks to the HPVM Guest, not the mirrored LV."

Very sorry, but it seems to be obvious, no ? Unless you want to do a mirror above a mirror ? That is said, why not : belt, strap and parachute ... Well, just a joke, don't care ;-)



And the question behind the question : what about massive NFS IOs above virtualization ? But it has nothing to do in this post. Unfortunately I will have to answer this question in a few weeks ... and I guess, Dave, you will be a good writer around this subject


Eric

[Dave Olker]

"the recommendation remains to do the LVM mirroring on the VM host "

The question behind this recommendation is : why. After all is it supported or not ?

- Performance reasons : datas have to go through the virtualization layers twice ?

- Virtual IO, whatever emulated scsi or avio, do not handle correctly hardware events ?

- Other ?
<<<<<<<<<<<<<<<<<<

Performance is the main reason. The performance tests we've done at HP indicate HPVM guests see the best storage performance when the host presents whole-disk LUNs. Next level down is logical volumes. Last is a file-based backing store. Reason is as you suspect, the less layers you have to go through on the VM host (storage, volume manager, VxFS) the better.

>>>>>>>>>>>>>>>>>>

Doing "LVM mirroring from inside the HPVM Guest then the HPVM Host needs to present the whole disks to the HPVM Guest, not the mirrored LV."

Very sorry, but it seems to be obvious, no ? Unless you want to do a mirror above a mirror ? That is said, why not : belt, strap and parachute ... Well, just a joke, don't care ;-)

<<<<<<<<<<<<<<<<<

Earlier this week I was working with a customer in Europe that was trying to setup LVM mirroring inside the VM guest and he was presenting the guest an LVM-mirrored logical volume from the host.

One thing I've learned doing support for 17 years: What's obvious to one is not necessarily obvious to all.

>>>>>>>>>>>>>>>>>>

And the question behind the question : what about massive NFS IOs above virtualization ? But it has nothing to do in this post. Unfortunately I will have to answer this question in a few weeks ... and I guess, Dave, you will be a good writer around this subject

<<<<<<<<<<<<<<<<<

Now you've got me curious. ;)

Are you thinking of storing the HPVM boot disks on an NFS server or having the HPVM guest act as an NFS client and mount NFS filesystems directly from the server?

One of those configurations is supported, the other is currently not. Hopefully it's obvious which is which.

Dave

[Eric SAUBIGNAC]

Well, I am supposed to go to bed now, it's a bit late here ... :-)

About performances and mirror inside a guest : one of my client, starting with HPVM 2.0, has LVM mirror inside the guest. At this time it was not clearly said that LVM was not supported in this configuration. This client works now with HPVM 3.0, LVM mirror still inside the guests. It is an unsupported configuration, but he has no performance trouble. Well, oracle handy home applications, and no heavy load. This could explains ... Of course, my others clients don't work this way.


About abvious things : nothing to say. You are perfectly right. You said in europe ? I hope he was not a french ;-)


About NFS ... Curious ... Well, in fact it is much simplier than you think : we have just to move some Netapp's HP-UX physical clients to HPVM guest. I have heard that NFS' IO are really heavy in this configuration ... So I wonder : virtualization + HP-UX NFS stack ... Oups. I guess you are angry ;-)


Well, I will jump now in my bed. See you later.


Eric

[David Dilly]

Hello Guys
Thanks for your reply.

But currently I still do not know what options to choose .....

A) Implement the mirror on the two storage arrays inside the VM Guests then present the whole disks to the VM Guests through the VM Hosts.

(This solution is the VM Guests that manages the mirror, but entire disks that are presented to the VM Guest).

B) Set the VG using the "shared" option and configure the mirror between the two storage arrays on the VM Hosts, and present logical volumes to the VM Guests.

(This solution is the VM Host that manages the mirror, but they are logical volumes presented to the VM Guest).

What solutions are supported and what is the most effective ?

Thanks again.

[Eric SAUBIGNAC]

Bonjour David,


Maybe a short extract from the "HP Integrity Virtual Machines 4.2 - Installation, Configuration, and Administration", page 181, 10.3.4, "Restrictions and Limitations of Online VM Migration" :

---------------------
Only whole disk backing storage consisting of SAN LUNs, and ejected file-backed DVDs, are supported for guest storage if you plan to migrate the guest online. File and logical volume backing storage are not supported for online guest migration.
---------------------

and

---------------------
NOTE: Integrity VM supports SLVM backing storage for online migrations. For details about shared LVM (SLVM) storage, see Section 11.5 (page 200).
---------------------

and my first post about support of LVM Mirror in a VM Guest


==> First answer : the 2 solutions you have described are supported.


Wich one is the most effective ....

Well, in fact I have really good performances with some clients that are configured with mirrored logical volumes at the host level.

To achieve this I try to avoid having a double LVM configuration (one at the host level and a second one in the guest) I don't configure LVM in the guest. Only vg00 is configured with LVM in the guest. I directly use disks DSF as FS ... For exemple /dev/disk/disk123 is mounted on /usr/sap. And it works pretty well.

So in this case, SLVM is the answer. But you need MC/ServiceGuard.

On the other hand, only one of my client has today an LVM Mirror in the guest. As previously posted, he has no performance trouble.

But despite this fact, and the recent support (maybe too recent) of LVM Mirror inside a guest, I am still doubtful : performances, reliability ?


==> Second answer : with MC/ServiceGuard, I would choose SLVM, that is LVM at host level. Without MC/Serviveguard, you have no choice : LVM Mirror in the guest and raw devices as backing store.

Eric

[David Dilly]

Hi Eric,
I agree with you.

I will configure LVM mirror inside the VM, so I will present the whole disk (raw) to the VM.

It is unfortunate that SLVM does not work without MC/SG .......

Thank you again for your help.

David.
(Allez les bleus)