Expanding the size of a shadowset

[Carleen Nutter]

I have a 3-member shadowset comprised of stripped 18gb disks. I need to expand it.
I have a 4 node VMS cluster, VMS 7.3-2. My disks are attached via a pair of HSG80s thru a pair of SAN switches.
I've been reading about Dynamic Volume Expansion but am unclear about something.
Some of my web-surfing indicates that to enable DVE on a pre-VMS7.3-2 disk, I just need to dismount it, mount it privately and do the $set vol/limit command. Then dismount and remount normally. Other threads indicate that, in order to migrate in a larger physical volumne, the initial shadowset devices MUST have been initialized with the /SIZE qualifier.
Can someone verify that DVE can be enabled without initializing; and, if so, does each member of the shadowset need to be dismounted, mounted privately, $set vol/limit, etc.? And does each member of the shadowset need the "$set vol/lim" done at the same time? or can they be done over time, say as a member is dismounted for a backup operation.
I'm just weighing my options and trying to minimize application unavailablity or avoid it altogether if possible.
TIA - Carleen

[Jon Pinkley · Accepted solution]

You do not need to reinit.

dism dsa device (not the members)
mou/ov=id dsaxxx
set vol/limit dsaxxx
dism dsaxxx
mou/clus dsaxxx label

I would suggest you try using lda devices just so you can verify for yourself how things work with your version of VMS, etc.

[Jon Pinkley]

Oh, I didn't specify dism/cluster.

Before starting you must dism/CLUSTER the dsa device (my example didn't have /cluster)

[Hein van den Heuvel]

Our friend Robert Gezelter did a presenation (1512) at the 2007 HP Technology Forum in Las Vegas which extensively covers all details.
As suggested, he uses small LD as examples.

See:

http://www.rlgsc.com/hptechnologyforum/2007/1512.html

hth,
Hein.

[Jon Pinkley]

"And does each member of the shadowset need the "$set vol/lim" done at the same time? or can they be done over time, say as a member is dismounted for a backup operation."

You do this to the DSA device, which will write the changes to all 3 members of the shadowset set.

Anything you do to a member when it is not part of the shadowset is lost once it is returned to the shadowset, as the member is resynchronized to the state of the shadowset. Therefore, you can't do this a member at a time.

[Jon Pinkley]

Carleen,

Definitely have a look at Robert's presentation that Hein referenced.

Unfortunately, it does not cover what you asked about. It starts with "Set the Stage" init/limit...

Still, well worth reading and trying out the examples with LD devices. Also, the LDDRIVER kit (search forums for LDDRIVER) has some examples using shadowing.

[Carleen Nutter]

Jon and Hein,

Thanks for your replies. I did see an earlier version of Bob's presentation, but as Jon said, it sets the stage. I'm already into the third act.

Jon's confirmation that a $init isn't necessary and that the $set vol/limit is done to the DSA device is helpful (not necessarily good news, thou, since it will make the volume unavailable).

I'll have to figure out how to test all this - I have no test system and my system is a high-availability, production cluster.
A very recent and unexpected change in our operations has me scrambling.
I won't have a downtime window to make any changes for several weeks - so will keep the thread open in case anyone else has any comments.

Thanks again for taking the time to reply.

[Robert Brooks_1]

This is off-topic, but I feel that with clusters, shadowsets, and V7.3-2 in the picture here, I must put in a plug for Host-Based Minimerge (HBMM), which was first supported on V7.3-2 with patch kits. Any recent UPDATE kit would have the requisite bits.

If you've ever experienced the pain of a multi-hour merge, you'd be quite amazed at a merge that takes just a few seconds!

-- Rob (a member of the HBMM engineering team)

[Robert Gezelter]

Carleen,

First, I would like to thank Hein and Jon for the reference to the session, it is appreciated.

This shortcoming was mentioned to me during the Technical Forum last week, but after the slides were already in could not easily be changed.

My first recommendation is to do testing and experimentation on LD devices. It is straightforward to install LD, and it is an excellent tool for testing procedures of this type.

I have been busy on several client matters since getting back from Las Vegas, but my information is that you can mount the volume privately and do a SET VOLUME/LIMIT.

Of course, the volume limit does not address the question of whether the Index file is of sufficient size.

This type of manipulation MUST be done to the entire shadowset as a whole entity. It cannot be done one member at a time (by definition).

If I can be of assistance, please let me know.

- Bob Gezelter, http://www.rlgsc.com

[Jon Pinkley]

Robert Brooks>>>This is off-topic, but I feel that with clusters, shadowsets, and V7.3-2 in the picture here, I must put in a plug for Host-Based Minimerge (HBMM), which was first supported on V7.3-2 with patch kits. Any recent UPDATE kit would have the requisite bits.

Rob,

Perhaps you could add a note to the "Disk to disk to tape" thread
http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=1136467

where I think it would be on be on topic.

[Jon Pinkley]

Robert Gezelter>>>Of course, the volume limit does not address the question of whether the Index file is of sufficient size.

Good point.

Carleen, at the time you do the set volume/limit you may also want to use the DFU 3.2 INDEXF command to defrag and extend the INDEXF.SYS file. I haven't tried this yet myself, and you will want to have a good backup, or split a member off before the change, perhaps using the write bitmap (HBMM) feature Rob Brooks mentioned, so adding the member back to the shadowset won't take a long time (just in case things don't work as planned).

One thing I don't see mentioned in the online help for DFU 3.2 is any way to increase the size of the header bitmap that is part of the INDEXF.SYS file. The size of the header bitmap is related to the /max_files. Now that Ton Dorland (author of DFU) is no longer supporting DFU, I am not sure how frequently it will be updated. It is being hosted by Jur (LDDRIVER) at http://www.digiater.nl

I don't know if the Defrag product (DFG) has an offline tool to modifiy max_files or not.

Jon

[The Brit]

Hi Carleen,
This does not address your question directly, however it does relate to the DVE issue.
After you have done the "set vol/limit", check the volume by doing a "show dev /full DSAnnn", and look at the Expansion Size Limit (which should be 1TB). Note this has nothing to do with your current expansion, but it does tell you what your maximum size can be for this DSA device (assuming that your LUN/Physical disk is big enough. 1TB is currently the VMS maximum volume size, and is the default if you dont provide an actual number with the "set vol /limit[=nnnnnn]" command

The Brit

[Carleen Nutter]

I have been using the mini-merge for awhile and Rob is right - it's amazing how quickly the merges take place. I will create some ld devices to test the process.

I may just take advantage of a downtime window and do a backup/restore to a new, larger shadowset (w/ $init/size, etc.) and get the disk completely defragged at the same time (most of our big files are always open, so on-line defraggers mostly defrag free space). Alot depends on how quickly I reach my freespace threshold.

It's always helpful to have options!

Thanks again everyone.
C.