HP-UX EMC PowerPath: power device or pvlinks?

[Dan Matlock_1]

Having a discussion with EMC right now and getting mixed signals on the whether to use the EMC powerpath power devices or regular pvlinks. Oddly enough EMC is stating to use regular /dev/dsk/c* devices and alternate links on HP-UX, so my question is whats the purpose of powerpath? We are new to EMC and tryin to implement...

[Geoff Wild · Accepted solution]

Powerpath provides automatic load balancing - pvlinks doesn't.

You can manually load balance with pvlinks - but then you are just guessing as to which path you want for the primary.

We use Powerpath on production systems - but not on QA/Dev.

Rgds...Geoff

[Christian Tremblay]

On some occasions, Powerpath may fail to switch to the alternate when the primary link
goes down. We did not have alternate pvlinks set up and found out the HARD way about that.

Don't assume that Powerpath will work, always set up an alternate link, that way if Powerpath fails, your a__ is covered.

Chris

[Dan Matlock_1]

Excellent responses, so I think the answer is going to be 'D' all the above!

We also are planning to use powerpath for prod., but not for our single attached dev servers.

EMC powerpath install docs show pvlinks and make reference to power devices not support on HP-UX. We have EMC intern workin here with us, and I questioned the docs....

I also wondered if ONLY using pvlinks then why did we buy powerpath, and where is our 'auto' balance loading?

So I will point to power device for main device and extend the VG to all paths for pvlinks....

[Scott Riley]

If PowerPath is installed and configured correctly, then it will fail over, whether or not you have the alternate path set up with PV Links.

To find out what devices are configured under PowerPath, issue a 'powermt display dev=all' command. Every volume in the array that has multiple paths will show those multiple paths in the powermt display listing.

PowerPath inserts itself in the driver stack, so using the /dev/dsk/cxtxdx device is fine, it will still load balance and failover appropriately. EMC used to require you to use the EMCPower devices, but they are moving away from that, so don't do it.

Be sure that you issue a 'powermt config' command after adding new devices to your system (after ioscan and insf). Then run powermt display to verify, and you are good to go. It's not a bad idea to ALSO use pvlinks, but it is certainly not required.

You can also use PowerPath to control load balancing and failover of other storage arrays, not just EMC. It works nicely with EVA and Hitachi-based arrays in my experience, and is a good solution for when you are migrating data between platforms.

The only time I have seen a properly configured PowerPath install fail to failover between paths is when an HBA took out a PCI bus, and both HBA's were on that bus...not exactly PowerPath's fault. However, I have seen pvlinks fail plenty of times. So best bet, just use both and you're covered.

[sathish kannan]

Dan,

PowerPath provide load balanvce. PV links will not load balance.

PowerPath loads its drivers and intercepts I/O. If you have PowerPath I would not recommend to use PV links. Addidng PV link is an overhead and some time it may create problems.

Operation systems apart from HP like SOLARIS, AIX etc uses Pseudo device names and HP does not.

You can change/display PowerPath options like failover, policy etc usning "powermt" commands.

Hope it helps.

Regards
Sathish

[Steve Post]

I was told by EMC specifically to NOT use PVlinks because it would mess up powerpath. It's in their manual. Ok I THINK it's in their manual. I read it last April (2006).

But I don't know what you mean by PVlinks. I'd assume you mean that you see this in vgdisplay -v
PV Name /dev/dsk/c3t5d6
PV Name /dev/dsk/c6t5d6 Alternate Link

Right?

[Rita C Workman]

Like everyone has said, PowerPath's role in your environment is that it enables load balancing. And it does an absolutely great job at doing it. PVLinks has it's function, but it's benefit is only felt when one link dies...it fails over to the alternate link. So they are a wonderful complement to one another.
PowerPath works great on HPUX and we use it on our Windows environment too. It has really helped improve I/O and performance.

Regards,
Rita

..in shorter terms, they are both good products to use..

[Dan Matlock_1]

Again thanks for the responses. Ya we figured out that unlike our redhat systems, I don't see a 'Pseudo device' name on HP-UX. EMC recommended that I just pick the first device listed on the powermt display dev=all and since we are not using meta devices on this clariion, we should not have to worry about picking devices on spa/spb = powerpath should manage. BUT they are telling me to include the 'alternate/pvlinks' to the VG as well. I think I'll put only one in and then yank the fiber and see what powerpath will do! Note: we have 4 paths from host to switch, and I want to ensure that if the one and only prim. device and its associated fiber card fails, powerpath and/or O/S will seemlessly switch for me.

[sathish kannan]

Dan,

Again I want to reitrate my statement that if you use EMC PowerPath do not use PV links. especially with CLARiiON being an active-passive array, adding PV link will cause havoc when the LUNS are tresppased.

So stick with PP and let it manage. As a procedure add the the default LUN owner as the device.

All the best.

Regard
sathihs