clustering: how to start?
zkoza at ift.uni.wroc.pl
Wed Dec 1 05:47:05 PST 2004
We have just bought 4 dual-processor Xeon boxes for a physics
department and want to turn it into a computing cluster.
This means we want that the users will communicate with it via
ssh/sftp and see all processors as if they were part of a one virtual
UNIX machine with 8 processors on board.
A typical job we want to run is
(a bunch of) numerical simulation programs that run for 2 weeks or so.
Thus, the cluster will serve only as a numerical workstation.
We have experimented with Debian and OpenMosix. The first outcome of our
tests is that Intels' so much advertised Hyperthreading technology is a
disaster. Then, OpenMosix seems to have a bug that can bring the whole
system to halt if one tries to run simultanieously several
We have also problems with compiling custom Linux kernel so that it
could see all the swap space and memory it should see.
Then I recalled that several years ago I experimented with FreeBSD and I
liked it. Now I hear that FreeBSD supports SMP "natively", so
no kernel recompilation will be necessary.
I thought - maybe we could try and use FreeBSD instead of Debian?
Now my question is: where should I start?
Which docs should I read?
Which program(s) should I use to cluster our 4 FreeBSD boxes?
More information about the freebsd-cluster