Construction and Application of an x86 Beowulf Cluster

1
Construction and Application of an x86 Beowulf
Cluster

• Susan Ditmore
• TJHSST Computer Systems Lab
• 2004-2005

2
Abstract
- Build a Beowulf cluster - 20-15 nodes - Support
for Pentium II and dual Celeron computers - Use
Openmosix - Use Systemimager - Integrate into the
lab
3
Background
- Summer of 2004 ousted many Pentium II MCWs
because there were now replacement computers -
MIPs cluster wasn't going anywhere - Many
transformers were broken but easily repaired -
There had been a cluster before, but it ran only
MPI. - Computers were stacked in the machine room
in hopes of resurrection
4
Description of Methods, Algorithms, Procedures,
and Process
What the cluster had to have - Automatic Load
Balancing (OpenMosix) - Automated Install and
Update (Systemimager) I wasn't going to net
install each and everyone, with the DS3 gone. -
Debian Linux (to ease use, it was to be as
similar as possible to workstations) - DHCP and
DNS, outside internet access - A cold environment
so all it's old hardware wouldn't die too
quickly - Restricted Access / Security
5
Description of Methods, Algorithms, Procedures,
and Process
What I did - Repaired and cleaned old Pentium
IIs. - Found a suitable place for the cluster
(this was harder than anticipated). - Created a
vanilla kernel with appropriate OpenMosix patch
(22 tries). - Installed and tested distribution
on Akosmia, golden client. - Tested and setup
router (DHCP3, IP Masq) - Network cluster (this
includes obtaining mac addresses, which was also
more difficult than expected). - Realized that
routing killed OpenMosix, because it uses the
default gateway. - Realized I needed OpenMosix
File System, and made that work - Setup a virtual
LAN through King. - Convinced the Admins to
implement Systemimager for the cluster - Setup
two test nodes while I waited, fiddled with
these - Realized that the virtual LAN worked, but
Systemimager would not work with Reiserfs. .
6
(No Transcript)
7
Description of Methods, Algorithms, Procedures,
and Process
What I must do - Implement Etherboot and PXE -
Examine the extent of the damage of the
powersurge - Release distribution to all nodes. -
Test cluster with various high-performances
programs. - Fix any potential issues and revise
image. - Secure cluster and attach it to the
network. - Create account forms and commence
more heavy duty testing. - Introduce Mr. Latimer
to the cluster. - Install MPI on the cluster. -
Find someone to the maintain the cluster the
following year Menyoung and Victor
8
Results and Conclusion

• Nothing is every easy
• Your project will take three times longer than
  you think it will take
• Openmosix will not migrate to a pentium II from a
  dual celeron

9
(No Transcript)
10
References
http//sourceforge.openmosix.net/ http//www.syste
mimager.org/ http//www.debian.org/