HP-UX

This may seem like an odd question, but I've heard that on old Alpha servers running OpenVMS, you could pause the system so that the OS is essentially suspended for a small period of time, then unpause it and it would pick up where it left off. During the pause, all CPU cycles would be halted, all i/o would be essentially dead. So I'm wondering if it's possible to do something like this on HP-UX 11.11 and 11.23.

My goal is to stop the systems from accessing or changing data on SAN hosted file systems long enough for me to get snapshots and then unpause the systems again. This would minimize our downtime for snapshots (to use for backup to tape). Since the file systems are on a SAN, the pause would not affect them. The reason I need to make sure that NO I/O happens is that all of the SAN hosted file systems are striped using LVM (for load balancing across the dual HBA and fiber paths to the SAN). I suspect that if even the most minute I/O happens while trying to make a snapshot in serial succession that the stripes would be out of sync and I'd have hosed snapshots.

So is there any way to do this, or am I dreaming?

What snapshots are you talking about and who is to do the backup?
But to answer your first question: Not that I know of... you would find that on some workstations (because at inactivity time limit, it dumps the RAM to disk and "pause"s).

Snapshots on an HP Storageworks EVA4000 SAN. Our general processing works like this:

1. Stop application server (with file system on SAN)
2. Stop Oracle DB (with data, archive logs, redo logs and application on SAN)
3. Unmount all of the above mention file systems (they are striped at the OS level to pull off load balancing over dual fiber paths, so there are double the LUNs on the SAN)
4. Make snapshots of all the LUNs
5. Mount, start up Oracle DB and Application server again

That's a rough approximation of what we used to do on the old SAN, now that we've moved to the EVA4000 I'm rebuilding things, so it occurred to me that we might be able to cut the downtime way down if we could either snapshot while the systems are still up. Someone suggested that Oracle hot backup mode would allow this on the Oracle side. The app server side is really where our biggest hit in downtime comes from since shutdown and startup of the services can take 10-12 minutes in total.

On the HP ITRC forums I was told pretty much that it seems like we're stuck because of the way the app server is designed (not my call to judge that and we don't own the code). So my thoughts now are to do a multisnap of all the LUNs with Oracle in hot backup mode briefly, and while the application server is up but at a time when it's likely to not be busy. Then just accept the fact that I'll need to run an fsck on the snapshotted file systems since they're still mounted when I snap them.

I'm still testing and have run into other problems which I will be posting about next. So I think we'll either have to live with the downtime or if making the snapshots while the systems are live but not too busy, we'll have to live with fsck.

If you're using LVM, you can do "vgchange -Q" to quiesce a volume group. But this doesn't mean that you get self-consistent snapshot. Suppose that a transaction is is progress... you might quiesce the volume group in the middle of the transaction. This objection appies to the pausing technique too.