From owner-freebsd-hackers Wed Mar 4 04:40:18 1998 Return-Path: Received: (from majordom@localhost) by hub.freebsd.org (8.8.8/8.8.8) id EAA27298 for freebsd-hackers-outgoing; Wed, 4 Mar 1998 04:40:18 -0800 (PST) (envelope-from owner-freebsd-hackers@FreeBSD.ORG) Received: from walter.doc.ic.ac.uk (tpYeu0BJZL49v+9A5aGWWsvxLhebrXk6@walter.doc.ic.ac.uk [146.169.2.50]) by hub.freebsd.org (8.8.8/8.8.8) with SMTP id EAA27150 for ; Wed, 4 Mar 1998 04:39:40 -0800 (PST) (envelope-from njs3@doc.ic.ac.uk) Received: from heron.doc.ic.ac.uk [146.169.2.31] ([BmchAoWCjGllqVVpcWQeLg7C8x3xuLkH]) by walter.doc.ic.ac.uk with smtp (Exim 1.62 #3) id 0yAD2R-0005Pu-00; Wed, 4 Mar 1998 12:12:43 +0000 Received: from oak2.doc.ic.ac.uk [146.169.17.7] ([3I3m0Zer1ew5CABu8AFvalyx24+D7ZzB]) by heron.doc.ic.ac.uk with smtp (Exim 1.62 #3) id 0yACrR-0006AL-00; Wed, 4 Mar 1998 12:01:22 +0000 Received: from njs3 by oak2.doc.ic.ac.uk with local (Exim 1.62 #3) id 0yACqr-0005la-00; Wed, 4 Mar 1998 12:00:45 +0000 From: njs3@doc.ic.ac.uk (Niall Smart) Date: Wed, 4 Mar 1998 12:00:44 +0000 In-Reply-To: Tom "Re: Cluster?" (Mar 3, 11:04pm) X-Mailer: Mail User's Shell (7.2.5 10/14/92) To: Tom , "Ron G. Minnich" Subject: Re: Cluster? Cc: Alex Povolotsky , hackers@FreeBSD.ORG Message-Id: Sender: owner-freebsd-hackers@FreeBSD.ORG Precedence: bulk X-Loop: FreeBSD.ORG On Mar 3, 11:04pm, Tom wrote: } Subject: Re: Cluster? > > On Mon, 2 Mar 1998, Ron G. Minnich wrote: > > > well for starters see http://www.sarnoff.com:8000/, go to metacomputing. > > > > clusters for freebsd have existed here since 1994. > > But that cluster design is focused on performance. The focus of this > dicussion is reliability. For example, being able to migrate functions > from a failed node to a functioning node with no user perceived losss of > service. I think a good approach to transparent clustering is through distributed shared memory. However, the coherencey schemes used for high reliability (probably sequential consistency update-based) are different from those used for high performance (release or entry consistency, update or invalidate based). In addition, with high reliability systems you must cater for the case of atomic updates to a range of pages in the virtual shared memory space. (some varient of entry consistency might help here) For my final year project I'm working on a shared memory implementation for the Fujitsu AP3000 using entry consistency, when I graduate (if :)) I'd be more than interested in adding some DSM support to FreeBSD to support transparent fall-over clustering. Thats going to be a couple of months from now though. One particular thing that could benefit easily from this are DNS servers, other servers like mail and news wouldn't be so easy, because of the need for a reliable shared filesystem. Plus there is the problem of how to get clients of these servers to contact the redundant one in the event of a failure, I think someone has done something in this area using proxy arp... Niall To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-hackers" in the body of the message