SAN Migration

[Waqar Razi]

One of our client is migrating from DMX 850 to DMX 950. EMC is taking care of SAN migration. The SAN is presenting 10 LUNS namely emcpowera to emcpowerj and all of them are part of the same Volume Group namely VG01. We are using EMC power path. The OS versions is RHEL 3.6 and PowerPath version is 4.3.5 (build 2). Now EMC is saying when they migrate LUNS, it will change the Logical Device ID and probably the emc pseudo device name.

We will be doing vgexport before this migration start and then vgimport after the migration is complete.

I have a question here. Since the powerpath device names of the luns is changing, does that mess up the data integrity of the volume groups?

[Matti_Kurkela]

> Since the powerpath device names of the luns is changing, does that mess up the data integrity of the volume groups?

No. Linux LVM does not care about device names at all. When you run "vgscan" (or the system startup script runs it when the system is booting up), LVM checks every partition on every disk to find LVM PV signatures. This is how it finds the PVs. Then it looks at the VGIDs stored on each PV that is a member of a VG. If all the PVs making up a VG are available, the VG can be activated and used.

In a SAN migration situation, vgexport and vgimport are very important in HP-UX LVM, but Linux is different. (On HP-UX, vgexport produces a map file for the exported VG. On Linux, VG map files don't exist.)

On Linux, you should still unmount the filesystems and deactivate the VGs for the migration... but after the system can see the new LUNs (i.e. the new emcpower* devices) and the old ones are gone, just run "vgscan" and then activate the VG again.

Doing a vgexport/vgimport on Linux should not be harmful, but it isn't necessary either.

Note: the FC drivers and the associated infrastructure was somewhat immature in RHEL 3. It has been my experience that hot-adding new LUNs without booting the OS will not always be fully successful on RHEL 3.x. Sometimes new devices don't show up at all; sometimes only some of them do.

Rebooting the system or simply unloading & reloading the HBA driver has helped even the most stubborn driver to detect all FC configuration changes.

RedHat says only RHEL 5 and above will fully support hot-adding FC LUNs. I assume this means "with drivers included in the distribution": updated FC drivers from the HBA manufacturers might perform better.

http://kbase.redhat.com/faq/docs/DOC-3942

MK