Scheduling large parametric modelling experiments on a distributed meta-computer

Nimrod is a tool which makes it easy to parallelise and distribute large computational experiments based on the exploration of a range of parameterised scenarios. Using Nimrod, it is possible to specify and generate a parametric experiment, and then control the execution of the code across distributed computers. Nimrod has been applied to a range of application areas, including Bioinformatics, Operations Research, Electronic CAD, Ecological Modelling and Computer Movies. Nimrod was extremely successful at generating work, but it contained no mechanisms for scheduling the computation on the underlying resources. Consequently, users would not have any idea when an experiment might complete. We are currently building a new version of Nimrod, called Nimrod/G. Nimrod/G will integrate Nimrod job generation techniques with Globus, an international project which is building the underlying infrastructure for large meta-computing applications. Using Globus, it will be possible for Nimrod users to specify time and cost constraints on computational experiments. Globus provides mechanisms for estimating execution time and waiting delays when using networked queued supercomputers. Nimrod/G will then use these to schedule the work in a way which meets user specified deadlines and cost budgets. In this way, multiple Nimrod users can obtain a quality-of-service from the computational network.