Component-based grid programming. A case study on wavelets

Grid middleware allows linking heterogeneous servers to internet-spanning computational grids. Component architectures can abstract over many technical arrangements and provide a higher-level interface to middleware. Thus, \Higher-order components" (HOCs) require from the developer to provide only application level code, and low level aspects are transparent to the developer. Similarly, high level parallel programming libraries allow the same abstraction for the developer, but these may be constrained by the choice of the target platform. For instance, the Edinburgh skeleton library (eSkel) can outperform component systems in some applications, but in the current version the target platform must have an MPI runtime environment, and the application must be written in C. HOCs in contrast can be deployed to any web service container and can even be implemented using multiple programming languages. In experiments on a grid-like testbed, we have applied a HOC for the \lifting scheme" algorithm to a wavelet transform on raster images. This application is integrated with eSkel and shows that the exibility of a component can be combined with the performance of a native parallel library.