Scyld and fstab for Diskless slaves
newt at scyld.com
Tue Dec 19 09:44:30 PST 2000
On Tue, 19 Dec 2000, Chip Coldwell wrote:
> Let me see if I understand you correctly: each node initially boots
> from floppy and then goes through a sort of second-stage boot that
> sets up the ramdisk, etc., is that right? So a local floppy disk is
> required on every node?
> Have you ever tried doing this with PXE? Or does the PXE reliance on
> tftp (even multicast tftp) make it undesirable?
We can do this with PXE. Erik Hendriks wrote 2 kernel monte so that we
would not need to rely on Intel to solve our cluster booting problem.
Given the complexity of the PXE spec -- it's amazing that it works at all.
In addition, PXE is of no use if you want to boot a Myrinet-only cluster
or a Gigabit Ethernet-only cluster without PXE BIOSes for these adapters.
If you want to use PXE, I have had good luck with HPA's syslinux PXE
loader. Our beoboot tool can be gently coaxed into spitting out
seperate kernel and ramdisk images for this purpose.
Our upcoming beoboot can be explicity invoked to create these images for
you. Users of the preview release can hack beoboot to not delete the
interim initrd image (comment out line 316) to get this file for use with
Scyld Computing Corporation
More information about the Beowulf