Disk Shadow Copies

[VMS Support]

Question: Three member shadow set.
Two members at site one
Third member at site two.
10 miles between sites.

Three node cluster - Two at site one
One at site two.

(We run OpenVMS 7.2-1H1).

Remove two members from shadow set (Both at Site one) and leave off line for a few hours.
Add one of disks at site two into shadow set.
Now have two member shadow set with second member catching up via copy of data from site two full member of shadow set.

Shadow set after period of time now has two complete members.

Add third disk based at site one into shadow set. Shadow server is running on node at site one also. We have read weights set so that nodes at site one will try to read from disk at site one and nodes at site two will try to read from disk at site two.

Will shadow copy take data from disk at site two or local disk at site one ?

I.E. Does a shadow copy always take data from first member to be added to shadow set or disk nearest to disk being added (Site weighting set).

Kevin Raven

[Ian Miller.]

There is a calculation based on the site weighting and the disk I/O queue length so local disks are favoured

How are the disks connected to the VMS system?

[VMS Support]

Disks are hosted on HSG80 via SAN.
So in our case we would have:

Member one - Site One - HSG pair One
Member two - Site One - HSG pair two
Member three - Site Two - HSG pair Three

So if node one at site one was adding disk member two to shadow set ..reads should come from Memeber one. Unless member one had large I/O queue and in that case read would come from member three.

[Volker Halle]

Kevin,

here is a pointer to a very detailled presentation by Keith Parris (HP) about OpenVMS shadowing:

http://www2.openvms.org/kparris/s2003_volshad_perf.ppt

The SET DEVICE/COPY_SOURCE command (if already available in V7.2-1H1) may be of interest here, as it allows you to specifiy from which source device the data should be copied during a full shadow copy in a 3-member shadowset.

Volker.

[Robert Brooks_1]

Kevin wrote . . .

(We run OpenVMS 7.2-1H1)

----

Please upgrade to a supported version of OpenVMS Alpha. As you appear to have a disaster-tolerant cluster, you will greatly benefit from many shadowing enhancements that
are available in V7.3-2, with appropriate patch kits. Of particular interest to you will be Host-Based Minimerge (HBMM).


-- Rob

[Robert_Boyd]

I want to add a vote of support for Robert Brooks' suggestion.

I have a site with very similar configuration and we use HBMM to make a quick recovery of shadowsets after breaking them to do backups.

With HBMM I cut down the time that shadowset recovery used to take in hours down to around 15-20 minutes maximum.

Robert

[Robert Brooks_1]

Robert wrote . . .

I want to add a vote of support for Robert Brooks' suggestion.

I have a site with very similar configuration and we use HBMM to make a quick recovery of shadowsets after breaking them to do backups.

With HBMM I cut down the time that shadowset recovery used to take in hours down to around 15-20 minutes maximum.

-----

Not to rathole the discussion of the original problem, but my *guess* is that you are using Minicopy, not minimerge, to reconstitute your shadow sets. Merges tend to take place in response to system crashes. Minicopy is frequently used in a situation such as you described, where a shadow set is broken apart for a short time, such as a backup.

Minicopy was first released for V7.3 and then backported to V7.2-2.

-- Rob

[Robert_Boyd]

Oooops ... you're right Robert -- slight glitch there ... meant to say Minicopy -- I'll blame it on the fumes from construction materials in the computer room for the last few days. Formaldehyde from temporary pressed wood walls. It's either that or age ;-)

It would be a relatively simple upgrade from 7.2-1H1 to 7.2-2 just to pick up the Minicopy capability. Personally I prefer 7.3-2, but with 7.2-2 you don't even have to worry about relinking any applications.

Robert

[VMS Support]

Thanks for the pointers and notes with reference to disk preference during disk shadow copies. We are looking at upgrading to OpenVMS 7.3-2. Sooner the better I think.

Thanks
Kevin Raven