[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Building a NOW with Debian



Hi there,

Since there's a very low traffic on this list I thought it would be OK
if I send a long post with many questions ;-)

Sorry if some of them are not exactly Debian-beowulf related, but none
of the other Debian lists seemed appropriate.  And since I use Debian
I thought I'd ask here before I turn to the Beowulf list.

I've got a project in my hands which involves using an existing PC lab
(the undergrads PCs, which runs NT) as a Beowulf-like Linux cluster.
The machines dual-boot NT and Linux, and we make use of them when the
students are on vacation.  The project is mostly for "experimental"
purposes in parallel computing.  They're connected by lowly 10Mbit
ethernet, so we're hardly going to run any Grand Challenge stuff.

This brings up my first question: anybody knows of a "loadlin" type of
program for WindowsNT?  Or a way to hack LILO to decide what to boot
by looking at the clock?  (the CMOS clock I assume)  It's more or less
trivial to schedule reboots from one OS or the other, but then I'd
like to have control also on the bootloader.  The idea is to have
automatic (unattended) switching from one OS to the other (and back)
based on some programmed schedule.  I asked this some time ago on
c.o.l.setup and didn't get any answers.

Now for the real meat.  I've decided that I'm building the cluster as
a NOW, rather than a Beowulf cluster.  The idea behind this decision
is that since each machine has 1Gb of HD available, I might as well
build a NOW, and make it more versatile than a Beowulf.  Graduate
students may then use the lab under Linux, and do both parallel
computations and "workstation" stuff.  It would somehow duplicate the
NT lab, and would be a perfect working example to show the University
how they could save $$$ if they chose Linux instead.

So here's my question: where are the docs relating to building a
Debian-based network of workstations?  I checked the Debian
Administration Manual and it's all "work in progress".

To be a bit more precise: I chose Debian because in the long run it is
the best to administer on a single machine.  But when dealing with a
distributed network running NIS/NFS, how does one go about in order to
keep the benefits of apt/dpkg?  (like painless upgrades!)  I guess I'm
just looking for tips, howtos, and possible gotchas when using Debian
for a PC lab running NIS/NFS.


An overview of my planned network structure:  [comments welcome!]
--------------------------------------------

*  there are 16 identical machines, each one with 1Gb of HD available
for Linux.

*  all Debian installation/upgrades centralized on one machine. This
will be also the NIS server.

* the rest of the nodes copy their root filesystem (everything except
\usr and \home) from the server, probably using rsync.  The finer
details of the personalization for each node (IP & hostname, \var
hierarchy and such) will be taken care by scripts, adding maybe DHCP
in there too. I'll need some comments here.

* each node, including the server, will have a largish \home
partition. User accounts will be distributed, then cross-mounted with
NFS.  A subdir in \home\scratch will be world-writeable locally on
each node, for ``scratch'' parallel I/O in MPI or PVM jobs.  The
server would actually have a smaller \home partition hosting only a
\home\scratch directory, because it won't have room for other user
homedirs.

* some other client nodes might also have smaller \home partitions and
then have a \usr\local or \opt in order to install non-Debian stuff
(commercial compilers or what have you), and export it with NFS.

* since the machines have IPs on the Net (I can't change that), I'll
need to use ssh. I hope it works OK with MPI and PVM.

* I plan to use shadow passwords now. Do they work OK with NIS in
slink or potato?
---------------------------------------------


So you see that under this plan all Debian upgrades (with apt or dpkg)
would take place on the master server, and then I would take care of
the rest of the machines more or less by homebrew methods.  But is
there any other better way?

Cheers,

JL
=============================================================================
Jose L. Marín                                   jose@ma.hw.ac.uk
Dept of Maths                                   marin@wanda.unizar.es
Heriot-Watt University
Edinburgh EH14 4AS, U.K.
Phone: +44 131 451 3893
Fax: +44 131 451 3249

Former address:  Dept. de Física de la Materia Condensada
                 Facultad de Ciencias, Universidad de Zaragoza
                 50009 Zaragoza, SPAIN
=============================================================================



Reply to: