Re-scheduling invocations of services on RPC-based Grid

RPC-basedgrid infrastructures emphasize on the composition of services on a large number of computing resources. The key issue to reach high performance is to enable exploitation of parallelism on servicesinvocations and communications. Moreover, this process should be transparent to reuse legacy codes. In this paper we present HOMA, an IDL compiler and a run-time support for automatic detection of the parallelism of invocations and their data dependencies on a set of CORBA objects. On homogeneous architecture, HOMA is accompanied by a predictable cost model. In the case of an application with a small critical path, among p processors the speed up of HOMA versus CORBA is asymptotically O(p). The illustrations on a case study in computational chemistry validate our cost model.