Clusters and distributed computing can operate in a number of ways. At one
end of the spectrum are full-time dedicated clusters. Alternatively, you can
get distributed power on the cheap if you set up a part-time cluster using
spare cycles on networked desktops.
Condor makes clustering and distributed computing possible for budget-conscious enterprises.
Condor is suitable for all
types of clustering. It was originally developed for use on
non-dedicated desktop clusters. Not surprisingly, therefore, it is particularly well-suited to those. The
project aims to provide policies and mechanisms to support High Throughput
Computing on distributive resources.
Condor functions as a batch processing system, with a central server that accepts
and queues jobs, serving them out to compute nodes as appropriate. If you’re
using a desktop cluster, you can set policies so jobs run only on idle
machines, and thus can be paused or moved transparently when a machine stops being idle.
Condor is incredibly adaptable — almost any usage policy and preference can
be described — and very versatile, making it suitable for many types of
distributed system. It can also interact with remote Grid systems
using a Globus interface, which requires installing only the Condor-G client
interface, not the server as well.
Condor is straightforward to install and deploy, and the online documentation and support available is great.
Find out more at the many links on the Condor Web page.