Operating System - OpenVMS
cancel
Showing results for 
Search instead for 
Did you mean: 

using mount/bind in OpenVMS 8.3

 
Lucie
Frequent Advisor

using mount/bind in OpenVMS 8.3

I hope someone can help me. I know this is not encouraged to bind volumes but the oracle disk is huge and and we want to use the feature in RMAN to have several backup thread fullin at once.

I have a 3 volume bind disk
$sh dev dga206

Device Device Error Volume Free Trans Mnt
Name Status Count Label Space Count Cnt
$1$DGA206: (N058V4) Mounted 0 ORA_FLSHBCK1 851.38GB 1 4
$1$DGA207: (N058V4) Mounted 0 ORA_FLSHBCK2 899.23GB 1 4
$1$DGA208: (N058V4) Mounted 0 ORA_FLSHBCK3 832.85


I need to added another 1TB disk to the set

mount/cluster/noassist/bind=ORA_FLASHBCK1 $1$dga209: ORA_FLSHBCK4 ORA_FLSHBCK1
-VOLINSET, volume is already part of another volume set

I specified the volume set name ORA_FLASHBCK1
The new disk is dga209: and it was init with a label of ORA_FLSHBCK4 no logical name.

What am I doing wrong?

11 REPLIES 11
Robert Gezelter
Honored Contributor

Re: using mount/bind in OpenVMS 8.3

Lucie,

Bound volumes are fine, providing you use the right procedures.

I would suspect that something is wrong with how the volume was initialized.

I would suggest doing the experiment using LD disks (small ones are fine).

- Bob Gezelter, http://www.rlgsc.com
Lucie
Frequent Advisor

Re: using mount/bind in OpenVMS 8.3

Hi Bob,

The new disk dga209 was initialized correctly, in fact I re-init the new disk several things.

It created another volume set and make the disk as rvn1 of a new set.

The existing volume set is already mounted. I want to add another volume to the set. There is nothing on the new volume, i.e the directory structure that exists on the other disks. Should the directory structure exist before mounting to a existing volume set?
Hoff
Honored Contributor

Re: using mount/bind in OpenVMS 8.3

You could initialize everything to get this to go, but recognize that bound-volume sets are an old and somewhat dodgy way to get a larger disk. (Be sure to test your recovery paths here.)

And you're clearly working around the 1 TiB limit within the current ODS-2 and ODS-5 disk addressing, which implies you might want to ring up HP and see if you can get the status of the oft-discussed 2 TiB project that's listed in the roadmap for V8.4.

If you're just working with ginormous BACKUP savesets as the labels here could imply, you could use the sequential disk set that's discussed in the manuals. And compression.

If you're growing your storage requirements, do start the plan for splitting out your storage access to multiple spindles, as getting past 2 TiB is likely going to be a far larger engineering effort than getting from 1 TiB to 2 TiB addressing; those longword disk addresses are just all over the place.

Stephen Hoffman
HoffmanLabs LLC
Steve Reece_3
Trusted Contributor

Re: using mount/bind in OpenVMS 8.3

I'm not sure that what you're doing should actually work, given Hoff's comments and my own experiences about the 1TB limit on disks on VMS. Doesn't a bound volume set still obey the 1TB limit?

Until v8.4 is released, the maximum disk size is 1TB and the maximum size for a single file is 0.5TB. This is why you can't create a disk bigger than 0.5TB using the SW-RAID product.

v8.4 of OpenVMS is promised to raise the limit on both of these but that ain't released until the middle of 2010.
Hoff
Honored Contributor

Re: using mount/bind in OpenVMS 8.3

The OpenVMS Bound Volume Set (BVS) is how you can get past the existing one terabyte limit of the ODS-2 and ODS-5 disk volumes. Within a single BVS, you can have up to 255 one terabyte spindles aggregated, in a theoretically maximally-hairy BVS.

For gonzo-level details on this file system stuff, see Kirby McCoy's VMS File System Internals book and/or the ODS2 spec that's on the Freeware.
Lucie
Frequent Advisor

Re: using mount/bind in OpenVMS 8.3

Until V8.4 is release and tested here, I am in a quandary. I need to find a way to add a volume to the set. I was hoping not to go down the route and re-init everything, but that may be the only solution.
When running Oracle RMAN backup on a single thread, it eventually crashes the OpenVMS systems. We tried to set RMAN to run 3 threads but that also crashed the system. Our thinking is if we add another volume we could run RMAN on 4 threads and hopefully not crash the system, this of course is theory.
Hoff
Honored Contributor

Re: using mount/bind in OpenVMS 8.3

Crashes? Forget the BVS. Something is badly wrong with either Oracle or OpenVMS here. Get Oracle patched (or upgraded) to current and get OpenVMS patched (or upgraded) to current; work to resolve whatever nasty is causing OpenVMS or Oracle to crash.

Get the crash fixed. That may involve direct contact with support for either HP or Oracle, or both. The CLUE CRASH data from one of these cases might point to a trigger; ANALYZE /CRASHDUMP with one of the dumpfiles and then CLUE CRASH. But get support involved.

Or migrate off of whatever is failing here. Folks can tend to inexplicably look at me funny when I suggest this, but if "it" is not working for you, then replace "it" with something that works better. Just because everybody always uses "it" doesn't mean there aren't also (good) alternatives.

And if you're not going to forget the BVS here, you can try to (re)initialize (just) the volume you're adding.
Lucie
Frequent Advisor

Re: using mount/bind in OpenVMS 8.3

Steve,
It is reported to HP and I have someone currently working on the crash. They reported the system crashed - oracle process was creating a directory.

Oracle version 10gr2 but I am not sure of the patch level. I will be putting the current OpenVMS 8.3 alpha and I64 patches on the weekend. Could adding a volume to a set be related to the crash?
Hoff
Honored Contributor

Re: using mount/bind in OpenVMS 8.3

>Oracle version 10gr2 but I am not sure of the patch level. I will be putting the current OpenVMS 8.3 alpha and I64 patches on the weekend.

The patches are usually the first thing HP wants done; that stuff is near the top of the standard support script.

>Could adding a volume to a set be related to the crash?

Without a look at the CLUE CRASH or at the crashdump, any speculation around a trigger seems premature.

It's taken me years to get some of the end-user folks I work and support with dragged away from the "What am I doing wrong" assumption, but I (philosophically) digress.

Regardless of whether this is a case of "What am I doing wrong?" (which, yes, can happen), if this is something that's not understood or problems in the user interface or the application affordance, or something OpenVMS or Oracle is doing wrong here, a system crash is aways bad and (unless it's your code that has directly triggered the crash) it is Not Your Fault.

An application or a system crash is typically not the fault of an end-user. It's typically the fault of a programmer, or of the hardware, or both.
Lucie
Frequent Advisor

Re: using mount/bind in OpenVMS 8.3

Steve,
Thank you for your sound advice. I have a feeling it will still crash after the VMS update. The job has been running fine until September. The oracle developers have finally loaded the database and the job was taking several days to complete. We made a change to RMAN to use 3 channel threads instead on 1. The system started to crash the change. We have since backed out of the change are running on a single channel but the job continues to crash the system.
I will take your suggestion and continue to work with HP engineers on the crash.
Lucie
Hoff
Honored Contributor

Re: using mount/bind in OpenVMS 8.3

> The oracle developers have finally loaded the database and the job was taking several days to complete

Several days?

Lacking rather more detail on the scale of the data and the hardware here beyond the obvious presence of Fibre Channel and a terabyte spindle or three, there can be no certainty here.

That written (and in the absence of a sufficiently large wad of data involved here), there might well be a database design issue here, or (and here's where DECset PCA or database-level monitoring can help) there looks to be a bottleneck lurking here somewhere.

Several days is a rather long run for an application. (And yes, I do know of stuff that runs days or weeks...)

Hopefully the database folks have implemented some sort of checkpointing here. And hooks for parallelism and/or database sharding and/or server upgrades are hopefully planned, within the constraints of the application requirements.