[Info-vax] Clusters, Quorum Disks, Shadowing/RAID-1 (was: Re: The Road to V9.0)

[Stephen Hoffman]

On 2019-07-09 07:36:42 +0000, Phillip Helbig (undress to reply said:

> In article <d8d756f6-2ff8-4f00-942a-6c8d2b287018 at googlegroups.com>, Jon 
> Pinkley <jon.pinkley at gmail.com> writes:
> 
>> Also, are there any initial plans about how shadowed system disks will 
>> be handled when the system disk is not the quorum disk?
> 
> Is there ever a situation where it would make sense to have the system  
> disk be the quorum disk?

Now?  Not really.  Not unless you're running hardware configurations 
with inadequate storage and/or inadequate connectivity, and of course 
nobody ever tries that.  Nobody ever gets "stuck" on old software and 
hardware, either.

That configuration was fairly common when disks were expensive—RA81 
disks were priced at US$12K—and somebody wanted the cluster to degrade 
gracefully.   Folks also usually wanted every byte of storage and every 
I/O on their data disks for their apps, and there was usually some 
capacity on a system disk.  (Yes, there were routinely discussions 
around quorum disk I/O, and one of the "fun" side-effects of sharing a 
quorum disk with data was that apps could generate so much I/O that the 
cluster could burble,)

One of various examples of this configuration: CI or DSSI cluster with 
three hosts and no "spare" disks, and a desire to keep the core cluster 
going with just one host and the system disk.  Whether for processing, 
or to allow satellites or NI-connected hosts to continue to operate.  
Usual here would be one vote to each of the three hosts, two votes to 
the quorum disk, Expected to 5, quorum of 3, which can survive with the 
disk and with any single host of the three.

Not that depending on a single RA81 was entirely reliable approach back 
then, but it was common. And this configuration could survive the RA81 
failure, save for that disk also being the system disk of course. And 
if you had the spare disks and had an HSC, you'd probably RAID-1 the 
system disk, and that also protected the co-resident quorum disk 
configuration.

With disk prices and storage configurations and controller-based RAID 
support where that all is now, this particular cluster configuration 
makes rather less sense.

Combined with increased server speeds and with increased hardware 
reliability, clusters are tending toward smaller, too.

> Why is the question of shadowed system disks related to the question of 
> a quorum disk?

A quorum disk cannot be configured with host-based volume shadowing; 
multi-host software RAID-1.  Were that permitted, there'd be the 
potential for several quorum disks to exist. That would be bad.  
Configuring a quorum disk with controller-based RAID-1 is fine.  This 
is a "Highlander" configuration requirement; "there can be only one."

-- 
Pure Personal Opinion | HoffmanLabs LLC