Diskless cluster on our existing computers
Szalai László
<>
NYME INGA
Edlinger Dávid
<>
NYME INGA
We frequently
have to solve computationally hard tasks with our IT infrastructure. A
good solution for these kind of problems is to build an array of
machines, often called a cluster. We connect our given hardware
according to a method or a system and the array as a whole does the
work.
Clusters are used for business and academic purposes too.
Today
anyone can build a multi-computer cluster using freely available
software components which is capable of sequential problem solving and
with a little ingenuity even parallel processing.
These clusters
cannot compare computationally with a supercomputer but in a
multicomputer environment they can give extra resources and most of them
are dual core and have great amount of memory and unused disk space.
In our
institute we have a huge collection of desktop computers, mainly in the
Computer-Studies labs and instructors' computers which have an
approximate weekly 25% utilisation. With the creation of the cluster our
main goal was to increase the utilisation of these computers without
changing the current operating architecture. The processing of heavy
duty tasks then can be done from the late afternoons to the early
mornings.
Other
objectives were to solve : group remote machine power on, booting from
network, user administration, choosing the cluster software and the
creation of the master and the diskless nodes. We had to create a shared
storage for the results and use the local disks for temporary
high-speed writes.
Everything was solved using a Linux based solution using only open source software.
An
extension was also created that enabled diskless virtual machines on
XCP hosts to be started remotely and for them to integrate into the
cluster that way increasing the number of nodes inside it.