home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.sys.hp
- Path: sparky!uunet!wupost!usc!rpi!batcomputer!cornell!alg
- From: alg@cs.cornell.edu (Anne Louise Gockel)
- Subject: Configuration for a Cluster of HP 720's
- Message-ID: <1992Jul28.194810.16039@cs.cornell.edu>
- Organization: Cornell Univ. CS Dept, Ithaca NY 14853
- Date: Tue, 28 Jul 1992 19:48:10 GMT
- Lines: 40
-
-
- I am configuring a lab that will have about 15 HP 720's in it. I've been
- weighing the options of configuring it as one or more clusters vs. using
- traditional NIS/NFS methods of managing a distributed lab. Right now I'm
- leaning towards the cluster approach.
-
- I would like this lab to have some redundancy in the servers. I can think of
- two options and neither is particularly clean:
-
- a) Have two clusters, a primary server on each cluster and split the clients.
- If one server goes down, the clients can be moved to the other server with
- minimal fuss.
-
- This does have the advantage of splitting the client load between two servers.
-
- However I believe I will have to use NIS and NFS to have a seamless user
- interface betwene the clusters. Right? Is there something I'm overlooking?
- Will the servers and clients cooperate with the NFS mounted /usr/spool/mail
- directory?
-
- b) Have a disk on one client that is in some sense a "hot spare". Basically
- this disk would be a copy of the server's disk. If the primary server goes
- down, this client can be brought up as the new server with minimal
- re-configuration. Will need to use "rdist" to keep most files updated and
- will need to pay particular attention to some of the cluster-specific files
- (/etc/clusterconf).
-
- Is there an "rdist" that works for HP-UX yet?
-
-
- Has anyone faced this decision and tried either option? Comments or
- suggestions (preferably based on experiences)? Is there some option that I am
- overlooking? I am getting the feeling that the "cluster" concept does not
- scale up to hundreds of machines particularly well, and I'm suspicious that
- there's some key ideas that I've overlooked.
-
- -Anne Louise Gockel
- Cornell Computer Science
-
- Internet: alg@cs.cornell.edu UUCP: cornell!alg
-