New clustered shadowset

[Martin Hughes]

> My current thinking is that during shutdown, the system should remove first the local drive, then the shadow set
<

That is right, you need to dismount the member first. Although I still think that the $ dismount $1$DKB200 command needs to be executed on one of the nodes that is staying up.

Also, I wouldn't dismount the DSA volume in SYSHUTDWN. I would just leave it to shutdown to dismount the volume, which it will do later in the shutdown process after having stopped processes and removed images etc.

[Jan van den Ende]

Martin wrote

>>>
Also, I wouldn't dismount the DSA volume in SYSHUTDWN. I would just leave it to shutdown to dismount the volume
<<<
Yes, if the shadowset is down to one member.
No, if there are still 2 members -- that would trigger a merge.
A clean shadow set dismount is just that: clean.
And the MINI version of COPY and of MERGE are orders of magnitude faster, still, they are not real cheap, with MERGE requiring 2 - 4 times as many IOs.

Proost.

Have one on me.

jpe

[Robert Brooks_1]

And the MINI version of COPY and of MERGE are orders of magnitude faster, still, they are not real cheap, with MERGE requiring 2 - 4 times as many IOs.

---

I'm not sure what you mean by this, but it's almost a given that you're wrong, with respect to HBMM.

HBMM only merges the range of LBN's that are known to have had writes since the last bitmap reset (or since the creation of the bitmap). Depending on the size of the bitmap, the duration of an HBMM-driven merge will likely be measured in seconds.

Perhaps our definitions of "cheap" are quite different, but by virtually any quantification, a host-based mini merge is
very "cheap".

-- Rob

[Aaron Sakovich]

Hi,

I thought I'd revisit this topic to be thorough and share my final (?) solution. At this time, this is what my mount and dismount routines look like. They seem to be working well based on my preliminary tests.

I have created a HBMM policy that allows all 3 servers to host the master lists.

At the very beginning of my SyLogicals.com on each system (NB, the alloclass varies by system):

$ Mount /System DSA10: /Shadow=($20$DKB200:) /Include -
/Policy=MiniCopy=Optional Common

In my SyShutdwn.com, it's a bit more complex. I first created a 'macro' of sorts to test if a drive is remotely mounted (see my blog post about MountCnt at http://openvms.hobby-site.com/pivot/entry.php?id=31 for details, but I hope to ultimately replace it with the new lexical arg MountCnt_Cluster someday); I then check to see if the system is running in a MIN startup mode, and finally use the macro to test if the drive is remotely mounted. If it is, I only dismount the local drive; if not, I dismount the volume.

$ TestRmtMnt == "Pipe Show device Disk$Common /full | " + -
"search sys$pipe ""Volume is also mounted on"""
:
:[various other commands...]
:
$ If f$edit(f$getsyi("Startup_P1"),"Trim") .nes. "" then Exit
$
$ TestRmtMnt
$ If $severity .eqs. "1"
$ then
$ Dismount Titan$DKB200: /Policy=MiniCopy=Optional /Cluster
$ else
$ Dismount Disk$Common: /Override=Checks /Policy=MiniCopy=Optional
$ endif