home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!think.com!mintaka.lcs.mit.edu!ai-lab!zurich.ai.mit.edu!gjr
- From: gjr@zurich.ai.mit.edu (Guillermo J. Rozas)
- Newsgroups: comp.sys.hp
- Subject: Re: clusters
- Date: 6 Jan 93 09:24:34
- Organization: M.I.T. Artificial Intelligence Lab.
- Lines: 39
- Message-ID: <GJR.93Jan6092434@chamarti.ai.mit.edu>
- References: <BzFsCM.H5u@bunyip.cc.uq.oz.au> <42692@sdcc12.ucsd.edu> <28226@dog.ee.lbl.gov>
- <TT.93Jan6150818@marius.jyu.fi>
- Reply-To: gjr@zurich.ai.mit.edu
- NNTP-Posting-Host: chamartin.ai.mit.edu
- In-reply-to: tt@marius.jyu.fi's message of Wed, 6 Jan 1993 13:08:18 GMT
-
- In article <TT.93Jan6150818@marius.jyu.fi> tt@marius.jyu.fi (Tapani Tarvainen) writes:
-
- | >We have ~15 300/400's and ~10 700s. Any significant comments on
- | >making a mixed-cluster out of these?
- |
- | I'd probably make two clusters of the lot -- two mixed clusters, not
- | one 700- and one 300/400 cluster (even though that would save disk space):
- | that way you can easily move machines from one cluster to another,
- | like when one server is down. But even a single 25-machine cluster
- | might not be a bad idea, depending on how they're used. You could
- | start with that and split it if the server chokes.
-
- We have a single cluster with 27 machines. A mix of 700s and 300s,
- although we are about to replace most of the remaining 300s with 715s.
- All our 700s have 48M+ and our 300s (except the couple of 320s left)
- have 32M of memory. Our 3 710s are diskless, everything else has
- local swap, but we are going to buy swapping disks for the 710s --
- they seem to perform noticeably worse than 720s with the same memory
- when they page, even though the lan and server are nowhere near
- saturated.
-
- The server is a 750 and is pretty idle, as is the LAN, but we tend to
- run compute & memory (not I/O) intensive software.
-
- It has been a very reliable setup (we used to have an 850 as the
- server before we had the 700s and that was a nightmare), and
- maintenance is much easier than when each machine had its own OS.
- |
- | >Was told HP plans to replace clusters with NFS in the near future.
- | >Is this true?
- |
- | Gack. I hope not. I'd much rather see the opposite, i.e., using the
- | HP diskless protocol for inter-cluster file access: it is faster, more
- | convenient with CDFs, and more reliable than NFS.
-
- I agree. I like the diskless stuff much better than any NFS-based
- stuff that I have seen. I just wish it weren't so sensitive to
- temporary network disconnect or server crash, although that is very
- rare (I think we've had one in the last year).
-