OpenSSI Clusters 1.2.0+ for KNOPPIX 3.6

[bjwatson]

Hello KNOPPIX fans,

I just finished doing an initial port of OpenSSI to KNOPPIX 3.6. For those not familiar with OpenSSI, it is a highly-available, comprehensive single-system image clustering OS, that allows you to manage your cluster like a single, giant SMP machine. All resources in the cluster, including filesystems, processes, devices, IPC objects, and more, can be transparently accessed from anywhere in the cluster. All you need to build an OpenSSI cluster are two or more computers and a private network switch.

There are a few caveats for this first cut of KNOPPIX with OpenSSI. One is that I deleted OpenOffice so that the compressed root image could fit on a 700MB CD. I don't think OpenSSI added an enormous amount of data to the root image, but I wasn't able to compress it small enough without deleting something big like OpenOffice. It can be restored in the future if the size of the root image can be reduced in a better way. Other known issues are listed below.

If you're interested in playing with this, you can download the ISO image from OpenSSI.org. Please let me know what you think, and feel free to copy OpenSSI's user's mailing list ([email protected]).

If you'd like to tinker with this integration of OpenSSI and KNOPPIX, and attempt to fix some of the issues below, so much the better! You can contact me with any questions or suggestions you might have. You can also feel free to copy OpenSSI's developer's mailing list ([email protected]).

Again, you can download the ISO image from OpenSSI.org.

Enjoy!

Brian Watson
OpenSSI project developer


Known issues:

- Administrative
* resolve .dpkg-new files that resulted from installing OpenSSI
* /var/run/utmp and /var/log/wtmp should be clusterwide files
* the kernel version number needs an OpenSSI revision number
* restore OpenOffice if space can be reduced in another way

- Debugging
* after starting X Windows, the mouse does not work
* when at runlevel 5, init does not start processes on node 2
* KNOPPIX fstab generation for local partitions is broken
* /var/run doesn't have everything from /KNOPPIX/var/run

- Development
* add "quickup" addnode feature, to avoid running ssi-addnode
* merge microcode support into devfsd

[mcreel]

Hey, this looks like a great project! I understand that LAM/MPI and MPICH will work with OpenSSI, correct? I have tried out the CD with a 2 node cluster. Bootup goes well. ssi-addnode is giving me a problem. The second node starts to boot via etherboot, but hangs with the message

.done
Issuing RESET:
mknbi-1.4.0/first32.c (ELF) (GPL)
Top of ramdisk is 0X0FFC0000
Ramdisk at 0X0FBE3000, size 0X003DD0000

This second node _will_ boot via Etherboot using the Knoppix terminal server, but it requires that the terminal server script include support for the via-rhine net card module. Could lack of this support be the problem here? I assume that ssi-addnode might need to know something about the type of netcard on the nodes, no?

Well, this looks like an interesting project. I'll be happy to help with testing if needed. Cheers, Michael

[bjwatson]

Hello Michael,

Yes, I think both LAM/MPI and MPICH work with OpenSSI, although I wasn't directly involved with this testing effort. IIUC, there's not much that OpenSSI does that would interfere with MPI. If you like, I can give you the e-mail address of the OpenSSI developer who did this testing.

This Etherboot bug has also been noticed by a few other users. It's due to an old version of mknbi that I accidentally included in the remastered image. I've heard that the latest version of Etherboot supports PXE emulation. You could try that, since I know I've successfully added nodes with PXE.

The ssi-addnode command does not need to know what kind of network cards are used on new nodes. Instead what I did was take every network driver provided by Fedora Core 1 (on which the current 2.4 OpenSSI kernel is based), put them in the initrd, and enhance the linuxrc script to try loading each driver until `ifconfig eth0' successfully returns a network interface.

You're more than welcome to continue helping with testing. I'm currently working on other things right now (like our 2.6 kernel port), but I'm keeping a list of all OpenSSI/Knoppix issues reported by users, so that they can be fixed the next time I or somebody else remasters one of these CDs.

Regards,

Brian