The
Nimrod/G project
The Nimrod/G project conducts research into metacomputing to provide access
to aggregated computational resources.
In particular, the project examines
-
parameterisation of serial programs to create embarassingly parallel
programs.
-
distributed scheduling to provide a uniform guaranteed completion time.
"Embarassingly parallel" programs split into independent tasks quite simply.
Distribution of these tasks is relatively simple since the tasks do not
communicate. Execution of multiple copies of a serial program can be seen
as such an embarassingly parallel task. Why run the same program multiple
times? Well, we may want to investigate what happens to the output if we
vary the input parameters to the program. Simulation programs are often
run like this, and program testing can be conducted in this way.
Once a user launches such a complex distributed program, they want some
control over its behaviour. In particular, they would like to know when
the entire computation will complete. Providing this information in a wide-area
distributed system is one of the challenges we intend to solve.
Nimrod/G is funded by the Distributed Systems Technology Centre (DSTC).
Current work is based at Monash University in Melbourne, Australia.
The Nimrod/G project evolved from an earlier project called Nimrod.
At the start of 1997, Active Tools released Clustor, a tool based on the
Nimrod prototype. This has since been acquired by TurboLinux.
Links