becker at scyld.com
Tue Dec 5 23:12:53 PST 2000
On Wed, 6 Dec 2000, LUK ShunTim wrote:
> > A common misperception when people first see the Scyld Beowulf system is
> > that it is based on a NFS root scheme.
> > Instead our model is based on a ramdisk root and cache, along with using
> > 'bproc' to migrate processes from a master node.
> 1. Does that mean that the scyld slave nodes can be started without any
> HD/NFS mounted root filesystem at all? And that's by all of the slave
Correct. Our CD is bootable. If the user doesn't select an master
installation within 30 seconds, the machine is booted as a normal slave node
using a ramdisk root. When booting from boot media the slave contacts the
master before accessing the local disk. Initial installation, disk
partitioning, checkpoint file recovery, and special filesystem or driver
modules are all handled consistently: the master node directs the slaves
individually or as a group.
We provide a way to permanently install the boot image on a local hard disk,
but the boot image is the same one as floppy, CD-ROM and PXE network boot.
> 2. What exactly are mounted in this root ramdisk? Is there any
> documentation that I can refer to?
The easiest was to find out is to try our distribution. The bulk of the
directory entries are in /dev/*. Most of the space is taken by /lib/*
(glibc is *huge*). We have eliminated most files in /etc, most importantly
/etc/passwd and /etc/group.
Donald Becker becker at scyld.com
Scyld Computing Corporation http://www.scyld.com
410 Severn Ave. Suite 210 Second Generation Beowulf Clusters
Annapolis MD 21403 410-990-9993
More information about the Beowulf