[Info-vax] HBVS (was: Re: Kittson question)

[Stephen Hoffman]

On 2015-01-08 08:48:43 +0000, Paul Sture said:

> Hmm, my memory says that I read similar in the V5.4 era, but maybe my
> memory is wrong.  Whatever, does anyone have more details on why Phase
> I was discontinued?

Shadowing Phase I presented DUS devices and was tied directly into the 
HSC storage controller, its firmware, and DSA/MSCP.

Shadowing Phase II used DSA devices, was entirely host-based, and 
solely required that the controller implement a few specific commands.

With the DSA/MSCP devices, the MSCP controller — specifically only the 
HSC series MSCP controllers, AFAIK/IIRC — would notify the host that 
the recovered sector data was corrupt every time the sector data was 
read; after the sector was replaced.  This notification allowed Phase I 
to detect and reload the questionable or corrupted sector from one of 
the other copies in the mirror.   Since that forced-error mechanism 
doesn't exist on newer devices, the (optional) READL and WRITEL SCSI 
commands are used.

If any member device involved doesn't have READL or WRITEL command 
support, then the disk volume gets kicked out of the shadowset on the 
first unrecoverable error.  This means that the first time there's any 
report of an error that's beyond the capabilities of the disk device 
ECC data, and thus that any device-local recovery could involve a 
latent disk sector corruption, the disk will be kicked out of the HBVS 
volume / RAID mirror set.

<https://web.archive.org/web/20071008224041/http://deathrow.vistech.net/~cvisors/DEC94MDS/quormq21.txt> 


More Google Fodder: HBVS / volume shadowing / host-based volume 
shadowing / RAID / Mirroring / EDC

Related: <http://labs.hoffmanlabs.com/node/838>

-- 
Pure Personal Opinion | HoffmanLabs LLC