Middleware Architecture for the Interconnection of Distributed and Parallel Systems

Abstract

Grid computing is a fast evolving technology, bringing more computing power to its users. Two main directions are observable: creating dedicated supercomputers for scientific and commercial tasks and creating distributed commodity-based systems. The first ones are usually much expensive, but have the advantage of performance, better control and uniformity in platforms. The second one is more affordable but lacks in flexibility and easy maintenance. The computing necessities that often require supplementary computing power for certain time periods are better satisfied by interconnecting available resources than buying new, expensive ones. But interconnecting platforms – sometimes radically different – can be a difficult task. The proliferation of hybrid parallel computing systems can be even more complicated because it puts in contact systems with various operating flows at the parallelism level. In this frame, the present article proposes a new middleware architecture that can connect multiple parallel or distributed resources, of different types, allowing unitary resource utilization and reservation for the user’s jobs. The new architecture is described functionally and structurally.