Thursday 4 June 11:00am
Location: Old Union - Clubhouse
June 1-5, 2009
|
|
|
Location: Old Union - Clubhouse
Speakers: Owen Le Blanc
Title: Sharing /usr across thousands of machines through AFS
Abstract:
The University asked me to look at providing a Linux service on more than 2300 cluster machines. Obviously installing and maintaining such a system could potentially be a large headache.
Currently we have a system which appears to satisfy our needs for the moment. The shape of the system is a little odd, since it makes use of facilities we already had, and it has to satisfy some of the peculiar demands of the decision makers. But it is easily installable, easily distributed, and easily maintained; for this reason it will perhaps interest others who are looking at similar requirements.
All the machine needs is an empty root partition and three files somewhere else -- a linux kernel, a ramdisk, and a config file. The ramdisk starts up networking and runs AFS, from which a minimal system is installed, about 130mb in size. If this already exists, its integrity is verified. Most of the linux installation is in AFS in read-only volumes. The system has now been in use for many years.
Slides: