Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!mnetor!seismo!lll-crg!nike!ucbcad!ucbvax!GRINNELL.MAILNET!McGuire_Ed
From: McGuire_Ed@GRINNELL.MAILNET
Newsgroups: mod.computers.vax
Subject: Re: Request for info on VAX cluster fail
Message-ID: <8607240938.AA11288@ucbvax.Berkeley.EDU>
Date: Wed, 23-Jul-86 13:26:00 EDT
Article-I.D.: ucbvax.8607240938.AA11288
Posted: Wed Jul 23 13:26:00 1986
Date-Received: Thu, 24-Jul-86 21:06:38 EDT
Sender: daemon@ucbvax.BERKELEY.EDU
Organization: The ARPA Internet
Lines: 23
Approved: info-vax@sri-kl.arpa

>Date: Mon, 21 Jul 86 10:12:03 -0500
>From: Mark H. Granoff <mhg@mitre-bedford.ARPA>
>
>I know of one installation that went from one 780 to a cluster
>consisting of a 780 and two 750's.  Before they clustered, the 780
>would crash at least once a week.  Since clustering, they have had
>almost no crashes.

I would almost like to think that the poor lonely 780 stopped crashing
because they gave it some friends to talk to.

An important thing about clusters is that if a node crashes the users
aren't transparently switched to another node.  For one thing, their
terminal line is connected to the node that crashed.  However, the user
can hopefully run to a different terminal and log onto a working node
and keep working.  If you give people a terminal server (or even just a
cable to each VAX and a rotary switch :-( ) then they don't even have to
run across the room.

DEC has been talking about (and hopefully doing something about) moving
processes to surviving VAXen in the event of a crash. I would like to
think that real process checkpointing might show up and make this
possible in the future.