Filesystem question (sort of newbie)

Jeff Layton jeffrey.b.layton at
Fri Oct 3 08:59:52 EDT 2003

Mark Hahn wrote:

> > 8 or but I haven't tested it), but it boots into a small Ram
> > Disk (about 70 megs depending upon what you need on the
> alternately, it's almost trivial to PXE boot nodes, mount a simple
> root FS from a server/master, and use the local disk, if any, for
> swap and/or tmp.  one nice thing about this is that you can do it
> with any distribution you like - mine's RH8, for instance.
> personally, I prefer the nfs-root approach, probably because once
> you boot, you won't be wasting any ram with boot-only files.
> for a cluster of 48 nodes, there seems to be no drawback;
> for a much larger cluster, I expect all the boot-time traffic
> would be crippling, and you might want to use some kind of
> multicast to distribute a ramdisk image just once...

While I don't prefer the nfs-root approach, Warewulf
can do that as well (haven't tried it personally). What
kind of network do you use for the 48-node cluster?
Anybody else use the nfs-root approach?

The 70 megs used in the ram disk is pretty well thought
out. There are some basic things to boot the node, but
it also includes glibc and you can easily add MPICH,
LAM, Ganglia, SGE, etc. The developer has thought
out these packages very well so that only the pieces
of each of these packages that needs to be on the nodes
actually gets installed on the nodes. Very well thought

Oh, one other thing. The image that goes to the nodes
via TFTP (over PXE) is compressed so it's about half
the size of the final ram disk. This really helps cut
down on network traffic (even works over my poor
rtl8139 network).

One of the things I'd like to experiment with is using
squasfs to reduce the size of the ram disk. IMHO, 70
megs is not very big, but reducing it to 30-40 Megs
might be worth the effort.

> regards, mark hahn.



Dr. Jeff Layton
Senior Chart Monkey - Aerodynamics and CFD
Lockheed-Martin Aeronautical Company - Marietta

Beowulf mailing list, Beowulf at
To change your subscription (digest mode or unsubscribe) visit

More information about the Beowulf mailing list