Distributed Computing is rapidly gaining importance with the advent of Web Services [1] and grid computing paradigms [2]. Java-based frameworks in particular, are receiving special attention due to their portability and interoperability. The growing synergy between the two types of technology can make possible a new generation of distributed computing frameworks based on independently written software components. Actual grid systems are typically tied to message passing interface such as MPI [3] and PVM [4] or to proprietary methods that impose strong limitations to the level of interoperability with other systems and other technologies. Web Services are based on a widely spread and standardized set of communication technology like XML and SOAP-HTTP. Therefore, Web Services technology seems to be a good way to overcome the interoperability problems typical of actual grid systems. However, the adoption of the standards fostered by Web Services for distributed scientific computing is limited to ancillary services and has not penetrated the core computational model of grid and meta computing. The motivation stays in the different community where the technologies have been developed: the scientific community for grid systems and the business community for Web Services. Web Services are oriented to transactions involving little amount of data where SOAP [5] is the standard communication protocol used. This protocol allows a platform independent and XML-based information encoding. However, such an encoding is extremely inefficient. In a scientific scenario big amounts of data are involved in the computation, typically in the form of numeric arrays. The use of SOAP introduces a huge overhead due to its XML information encoding. It is our opinion that this issue is very important. In fact, we think that the adoption of Web Services inside the computational core of grid computing applications is totally unfeasible as long as this problem has not been solved. In order to make Web Services standards well-suited to the requirements of high performance computing some modifications are needed, and it is necessary to exploit the extensibility of Web Services technologies. In this paper we provide a technological overview and an investigation on the performance result of different Java based Web Services implementation. We consider the actual de-facto Web Services standard framework, namely AXIS [6] that is provided by Apache Software Foundation and two frameworks that aim at providing Web Services-like communication methods oriented to high performance computing: the XSOAP [7] toolkit provided by the University of Indiana and the WSDL Grid Binding [8], a WSIF [9] extension provided by the University of Genoa. This paper is structured as follows: in section 2 we provide a general overview of the Web Services standard technologies, in section 3 we describe how the three infrastructures implement those standards, in section 4 we show our performances analysis and in section 5 we provide some concluding remarks.
展开▼
