Technology - Sustainable multi-scale simulations using Grid Remote Procedure Call
Scientific grid environments often rely on compute resources of varying capacity-scattered across multifarious locations-to process problems of miscellaneous size. So how can you write an application program that allows you to make effective use of such distributed computing resources, especially over months or even years?
One way is to ensure your programming model is flexible, scalable and fault-tolerant.
Your application must be able to request additional computing resources on-the-fly and according to availability; effectively manage a large number of parallel activities; and automatically recover from cluster node or interconnection failures.
A solution that meets these critical requirements is Grid Remote Procedure Call, or GridRPC, an extension of standard RPC tailored to the grid environment.
Simple, powerful, practical
Conventional RPC allows a computer to execute a procedure at a remote address, without requiring specific instructions detailing this interaction.
When viewed at a very high abstraction level, the programming model provided by GridRPC is that of standard RPC, with the addition of asynchronous, coarse-grained parallel tasking.
At a more practical level, GridRPC provides a simple, yet powerful, client-server-based framework for programming on grids.
Easy access; growing potential
GridRPC is an Open Grid Forum standard that makes grids easily accessible for hundreds of users already applying the RPC model to their scientific applications.
The growing GridPRC user base already includes Ninf-G, developed by the National Institute of Advanced Industrial Science and Technology in Japan; GridSolve, developed by the University of Tennessee Knoxville in the U.S; and DIET, developed by Institut National de Recherche en Informatique et Automatique in France.
The potential of GridRPC has been most recently demonstrated at SC07 in October 2007, where it was used in a 60-hour simulation distributed across 1129 TeraGrid and AIST processors on the Trans-Pacific Grid infrastructure.
The simulation ran without any need for manual restart, automatically recovering from system faults. These and other experimental results indicate the strong potential for GridRPC in implementing sustainable grid applications.
The Open Grid Forum GridRPC working group is now tackling service discovery, introspection, persistent data, workflow management, and the proper expression of these same within GridRPC.
- Yoshio Tanaka, Open Grid Forum GridRPC Working Group Secretary