Parallel Kernels: An Architecture for Distributed Parallel Computing

Global optimization problems can involve huge computational resources. The need to prepare, schedule and monitor hundreds of runs and interactively explore and analyze data is a challenging problem. Managing such a complex computational environment requires a sophisticated software framework which can distribute the computation on remote nodes hiding the complexity of the communication in such a way that scientist can concentrate on the details of computation. We present PARK, the computational job management framework being developed as a part of DANSE project, which will offer a simple, efficient and consistent user experience in a variety of heterogeneous environments from multi-core workstations to global Grid systems. PARK will provide a single environment for developing and testing algorithms locally and executing them on remote clusters, while providing user full access to their job history including their configuration and input/output. This paper will introduce the PARK philosophy, the PARK architecture and current and future strategy in the context of global optimization algorithms.