[Info-vax] HBMM vs HBVS

[Robert A. Brooks]

On 10/25/2011 4:03 PM, Kenneth Fairfield wrote:

>>  I wrote . . .

>> HBMM uses the Write Bitmap technology that first appeared in V7.3 (and
>> backported to V7.2-2). It was created to enable HBVS minicopy.

>
> And for a little history, folks I worked with at
> Intel (and I'm sure there were other customers as
> well) were very unhappy to find that shadow sets on
> FC-attached disks (Brocade SAN and HSGs) would undergo
> a FULL MERGE whenever there was a "bump" in the
> cluster (yes, usually a node crash).  They/we
> really missed minimerge as implemented on CI (HSJs
> and HSCs).  Intel brought a lot of pressure on
> Compaq to fix this omission.
>
> My recollection, now quite cloudy, of the initial
> design was to have special HSG firmware work with
> a reserved area on each shadow member which would
> function as the write bitmap.  That work went on
> and on and on over about a three year period (at
> least).

Yeah, that's largely correct.  The "reserved" area actually logged
the LBN and the block count; it wasn't a bitmap.

> At some point, whether during a review or at some
> implementation check-point, storage engineering
> determined the design just wouldn't fly and they
> dumped.

> However, by that time minicopy was in production
> and they realized they could leverage the minicopy
> bitmaps for minimerge.  It didn't take long after
> that to get minimerge out the door.

Not quite true.  The HSG80 "Write History Logging" (WHL) project
started after the Write Bitmap (WBM) project.  The WHL project
was behind the eightball from the beginning.  It really didn't have the 
backing of the storage group (the folks that wrote the HSG firmware). 
They were kind of dragging their feet, and not really thrilled about 
doing something VMS-specific.  Having said that, the biggest problem was 
in the shadowing driver; the paradigm wound up having a hole or two that 
should have been discovered during the investigation phase. 
Unfortunately, that discovery wasn't made until after a significant 
investment of time had been made.

Another issue with WHL is that it would have been HSG-only.  That
work would not have carried on to the HSZ, EVA series, or any 
third-party controller (obviously).  I think it's pretty clear that 
using the host-based solution was by far the best idea.

> We were on v7.3-1 at the time, but some small glitches, similar
> to things you found in the command syntax/parsing,
> weren't fixed until the v7.3-2 ECO came out.

Right; we did our initial work on V7.3-1 and released that version
to Intel only.

			-- Rob