Moving SANS/BCV

[M Trux]

I have 2 servers on the same EMC array.
1 server is 11.23 and the other is 11.31.
Can I move the SANs from one 11.23 to the 11.31 server and still use the data.
It is only data filesystems no programs.

Thanks

[Pete Randall]

Sure - take a look at vgexport/vgimport.


Pete

[M Trux]

So as long as I export them on the 11.23 server and import them on the 11.31 server it should work?

Does this include raw disk?

So is there no impact on moving from leagacy to Agile filesystems?

[Pete Randall]

>So is there no impact on moving from leagacy to Agile filesystems?

Good question. I'm assuming that HP, in their wisdom, provided some sort of migration path, but not having dealt with 11.31 yet, I'm afraid that I'll have to leave that question for others more familiar.


Pete

[Matti_Kurkela · Accepted solution]

If you have legacy filesystems that were originally created using HP-UX 11.00 or 11.11, you may have to upgrade their VxFS disk layout version to the 11.23 current level before the upgrade.

As a rule of thumb, as long as your filesystems have been created using the default layout version of your current OS version, you're upgrading the OS, and not skipping versions (e.g. going from 11.11 directly to 11.31), both the LVM and the VxFS filesystem will always have the necessary backwards compatibility.

If you're downgrading or skipping versions, you'll need to check more carefully.

This document identifies the various VxFS disk layout versions available in each HP-UX version, all the way back to HP-UX 10.01.
http://docs.hp.com/en/5900-0547/5900-0547.pdf

LVM is very easy when upgrading, but you may have to check the version-specific limits if you're downgrading for some reason:
http://docs.hp.com/en/6054/LVM_Limits_White_Paper_V4.pdf

The Agile "magic" happens at the disk device level, which is below the LVM level. As far as LVM is concerned, the Agile devices are ordinary disks with slightly different names. Of course, the native multipathing of Agile devices makes the LVM-level alternate path feature redundant.

For the Agile multipathing to work, the storage system must present the LUNs with unique identifiers in such a way that 11.31 can identify which paths lead to the same LUN. This may require some changes to LUN presentation settings (e.g. selecting "Host Type" as "HP-UX 11.31+" instead of just "HP-UX" for the HBAs connected to 11.31 systems). Make sure your SAN admin knows which HP-UX systems have 11.31 and which ones still use older versions.

When vgimporting using the -s option on 11.31, you should remember to use the -N option too, if you want the VG imported using Agile disk devices; if you forget it, the import will use legacy devices.

After vgimporting with -s -N options, check that you *don't* have alternate paths at the LVM level: if you do, it means the Agile multipathing system failed to identify the lunpaths as pointing to the same LUN, and it often means that the storage presentation configuration is incorrect for 11.31.

Raw devices are not a problem: even if a particular LUN shows up as /dev/rdsk/cXtYdZ on 11.23 and as /dev/rdisk/diskNN on 11.31 with the Agile naming, the application will see the exact same bits in the exact same order within the device.

----
We had to run one cluster for a while with one node at 11.23 while others had 11.31, but that caused no problems for Serviceguard. We had some package disks that mounted as legacy devices on 11.23 and Agile devices on 11.31 - no problems in failover either way.
(I know mixing OS versions within a cluster is officially supported for the duration of a rolling upgrade only - but yes, it works.)

MK

[M Trux]

Great.
Thanks for the in depth response.

[M Trux]

See above