From:	SMTP%"hasan@state.demon.co.uk" 12-MAY-1993 12:41:50.92
To:	EVERHART
CC:	
Subj:	Re: Pros and Cons of single system disc in DSSI VAXcluster 

X-Newsgroups: comp.os.vms
From: hasan@state.demon.co.uk (Hasan Ali)
Subject: Re: Pros and Cons of single system disc in DSSI VAXcluster 
Distribution: world
Organization: State Modules LTD
Reply-To: hasan@state.demon.co.uk
X-Mailer: Simple NEWS 1.90 (ka9q DIS 1.19)
Lines: 64
Date: Tue, 11 May 1993 13:22:36 +0000
Message-Id: <737126556snz@state.demon.co.uk>
Sender: usenet@demon.co.uk
To: Info-VAX@kl.sri.com
X-Gateway-Source-Info: USENET

In article <737071339snz@kestrel.demon.co.uk> ken@kestrel.demon.co.uk writes:

>I am about to set up our first dual host DSSI VAXcluster and I am looking for
>advice on configuration of system discs.  I'm sure this is a tiny system
>compared to the large clusters many of you are managing but I would be grateful
>for the benefit of your knowledge and experience.  
>
>I can think of three possibilities:
>
>1. A single system disc.  Easiest to manage but lose it and you lose the whole
>   cluster.  Also there must be a price to pay in performance having both
>   systems using the the same system disc.

I set up a system of this type back in November last year.  The system 
consisted of a couple of 4400s, 10 RF72s and an RF35 as the system disk.
We had some money floating around in my department that had to be spent so
I managed to solve the problem in a nice, but expensive, way - phase II 
volume shadowing and another RF35 - if you're an academic site though the 
license costs next to nothing (under DEC's all licenses for about 2000 
deal.)  There hasn't been any particularly significant improvement in
performance, but I have a system running a rather maverick Oracle based
application, and the IO queues on the database disks, outweigh everything 
else :-)  BTW this takes my system up to maximum DSSI capacity - if I wanted 
more disks I'd have to go SCSI adaptor (MTI do something like this that 
allows about 40 DSSI nodes!)

>What would the quorum setting be for these configurations? 

On my setup I have a disk designated as the 'quorum' disk, so 

QDSKVOTES=1
VOTES=1             ! on each node
EXPECTED_VOTES=3    ! THIS is the relevant one to you: QUORUM has been
                    ! replaced with EXPECTED_VOTES since VMS 5.0

EXPECTED_VOTES should be equal to the maximum number of votes in your 
cluster (sum of nodes votes + quorum disk)

>What are the pros and cons of having all storage in an expansion box separate
>from the two processor boxes?

I would say mostly pros - I have all my disks in two expansion boxes - both
coming off different power supplies (though same phase of course :-)).  If
I lose one power supply half the system keeps going; the disk mapping is
such that the remaining node should be able to continue to run unimpeded -
of course this wouldn't be possible without volume shadowing, or similar.

The cons are mainly cost - the boxes are pretty small and there shouldn't
be any problem in physically accomodating them.

>Most of DEC's discussion on this subject focuses on redundancy to protect
>against CPU failure.  I'm more concerned about the system disc being a 
>single point of failure which in my experience is more common (though I haven't
>had an RF disc fail yet...).  Have I got my priorities wrong?
>
>Ken 

CPU failure is unlikely (although I've recently heard of a combination of
CPU and memory failure bringing a 6640 to it's knees - DEC eventually did
a box swap); and disks *are* far more likely to fail, so I would say your
priorities are reasonably appropriate!

-- 
Hasan Ali