Storage works merging for 24 hours after node crash

[kendall Mccarthy]

We have a two node alpha cluster . In the past couple of days , one of the nodes has crashed twice. After it reboots, the shadow set is taking up to 24 hours to merge the shadow sets and peggs CPU on the nonchashed node. Any ideas?

[Robert Brooks_1]

What version of VMS? If it's V7.3-2 (with recent patch kits) or later, you should strongly consider using Host-Based Minimerge (HBMM).

If HBMM is not an option, please see the documentation regarding the logical name SHAD$MERGE_DELAY_FACTOR


-- Rob (one of the HBMM engineers)

[kendall Mccarthy]

OpenVMS V7.3-2 , Ive never seen shadow server take this long to complete the merge.

[Jan van den Ende · Accepted solution]

Kendall,

to start with:
WELCOME to the VMS Forum.

The time you observe for Shadow Merges has for a LONG time been an issue.
That is, it BECAME an issue when non-DSA-compliant disks came into wide use.
If you were shadowing devices on HSC, (and I beleive both of HSJ and HSD, no experience there, so not sure) you were used to shadow merge times of several SECONDS.

HBVS shadowing of SCSI devices however took MANY hours. Your 24 hour is no record, by far.

But, Engeneering FINALLY managed to get HBMM (HostBasedMiniMerge) to work. It certainly earned them a BIG applause from me (on behalve of our users)!

Obviously you do not have HBMM activited.
If ever there was an incentive to do just that, _YOU_ have found out the hard way!

Just instal HBMM_002 (I have not got the exact name, but you need V2).
The hardest part is finding the command in the realease notes, and you are all set.

Success!

Proost.

Have one on me.

jpe

[Ian Miller.]

There has been various issues with this sort of thing. Are you reasonably current on patches?
(esp shadowing, fibre_scsi)

[kendall Mccarthy]

I saw a note out there about hbmm being on hold for VMS 7.32. anything I should be concerned about?

[Robert Brooks_1]

With respect to HBMM being on hold for V7.3-2; that issue is almost two years old. HBMM for V7.3-2 was released in fall 2004; shortly after we released the kit, we found an issue that needed to be addressed. Any UPDATE kit from 2005 and beyond will contain the correct HBMM bits.


-- Rob

[Thomas Ritter]

We just completed Host-Based Minimerge (HBMM) deployment on all of our VMS 7.3-2 systems.
On the Test Cluster,after a crash, disk merges would take about 3 days to completed. After HBMM was introduced the merges were completed in about 15 minutes. Repeat 15 minutes....
Treat HBMM deployment as a project. You will need to know your disk usage patterns and configure accordingly.

[Robert Brooks_1]

15 minutes? What's the reset_threshold on the various policies?


-- Rob

[Jan van den Ende]

Like Robert, I am surprised at those 15 minutes.
We went down from 18 hours to well under 1 minute.

fwiw.

Proost.

Have one on me.

jpe