Physical Volume mounted by multiple hosts

[Enrico Venturi]

Dear all,
as in your knowledge, is there any products or any planned product which allows to multiple host to mount the same physical volume? My question arises because I'd like to have two applications, the main and the standby in a MC/SG environment, which share the data disks; when the main fails the spare is ready to run withount the need of the startup.
Best regards
Enrico

[Jeff Schussele]

Hi Ventruri,

Well, MC/SG will allow two hosts to mount same VG but NOT both as r/w. Only one can mount r/w & the other as r/o.
I'm sure you can imagine the bad things that can happen if 2 systems try to write (with buffers) to the same VGs at the same time......

Rgds,
Jeff

[John Poff]

Hi,

If your application is so critical that you can't be down for a few minutes during a failover, you probably need to be considering fault tolerant systems. Very expensive!

JP

[melvyn burnard]

Well the only real application like this is ServiceGuard OPS Edition, now renamed as ServiceGuard extensions for RAC, using Oracle's 9i RAC database, and which allows you to have the Database active on multiple nodes.
If your application is such that it does not use OPS or RAC, and you cannot tolerate the few minutes for the application to switch and startup, you are looking towards Fault Tolerant systems, and the price skyrockets.

[A. Clay Stephenson]

For filesystems mounted as read-write this cannot be done. You can actually get both machines to mount a filesystem rw but absolute chaos will result in no time.

Some products share raw volumes so that buffer cache is not involved and filesystem data structures are not in play.

With the exception products like Oracle RAC, forget about this plan.

[Martin Johnson]

You should check out the Veritas suite of products (http://veritas.com). Their cluster software coupled with the Veritas Volume Replicator can provide a hot spare capability.

Since HP now owns Compaq's Trucluster technology, you may also want to look into this also. It is based on DECs OpenVMS Cluster technology which has a distributed lock manager that allows read/write access from all nodes in a cluster.

HTH
Marty

[Enrico Venturi]

Colleagues,
all of your replies are really interesting!! Thanks a lot for your spent time.
Again: Jeff (Schussele), do you know when the contemporary mounting r/w and r/o will be allowed? Is it matter of MC/SG or of the Logical Volume Manager? I think that MC/SG isn't in charge of the multiple mounting ... it is matter of the pkysical volumes management, apart any application running on top of them (including MC/SG).

thanks again
Enrico

[Michael Lampi]

Ok, this is not an HP solution, but SGI IRIX has something called CXFS which does this sort of thing. It is being ported to Linux, with an expected release date sometime early next year (2003).

[Steve Lamont_1 · Accepted solution]

vgexport volume group from primary node

# vgexport -s -p -m -v /tmp/vg01.mymap ???f /tmp/vg01.out /dev/vg01

on second node vgimport volume group info
after copying mapfile and device info file to second node.

#vgimport -v -m /tmp/vg01.mapfile ???f /tmp/vg01.out /dev/vg01

1. Deactivate the volume group on primary node using
#vgchange -a n /dev/vgxxxxxx

2. Change the properties of the volume group using
vgchange -c y -S y /dev/vgxxxxxxx

3. Activate the volume group on both nodes using
vgchange -a s /dev/vgxxxxxxx

[Dietmar Konermann]

Sorry for reactivating this old thread... I just stumbled on it feel the need to comment.

It is NOT allowed to mount a standard HP-UX filesystem (VxFS or hfs) r/w from one side and r/o from another. Neither with SG nor with SG OPS edition. The r/o side simply wouldn't get any updates... instead it would get an inconsistent view caused by data caching. An, of course very limiting, option would be r/o mounting from all sides.

Other options could be cluster file systems (like advfs/cfs, which is planned to be ported from Tru64 to HP-UX).

Best regards...
Dietmar.

[Martin Johnson]

Dietmar,

DEC clustering by any other name...

:-)
Marty

[Pete Randall]

Dietmar, et all,

Fair warning from (unfortunate) experience: Not only is it possible to mount from both, it's even possible to mount r/w as I found out when I inadvertently trashed my production database because I believed the man page that said I couldn't activate the volume group on the second node. And this wasn't even a SG environment. Everyone is absolutely right when they talk about the chaos that can result. Thank heaven for good backups.

Pete