next up previous contents
Next: Current Status of HPJava Up: Future Works Previous: Java Numeric Working Group   Contents


Web Service Compilation (i.e. Grid Compilation)

One of our future research interests is in developing the Web Service Compilation (i.e. Grid Compilation) technique. As a good example of Grid Computing System, web services are the fundamental building blocks in the move to distributed computing on the Internet. A web service is any service that is available over the Internet, uses a standard XML messaging system, and is not tied to any one operating system or programming language. Basically, a web service protocol stack consists of SOAP (i.e. how to talk to web services), WSDL (i.e. how web services are described), and UDDI (i.e. how to find web services). There is a common feature between Parallel Computing (e.g. HPJava) and Grid Computing (e.g. Web Service). Both computing systems heavily depend upon messaging. Although explicit message-passing is a necessary evil for parallel computing, messaging is the natural architecture for Grid Computing. On the current scientific and engineering technique, the main difference for messaging between parallel computing and Grid Computing is the latency. For example, MPI's latency may be 10 microseconds. How can the message systems of Grid Computing catch up with MPI? There are some areas we can research to improve the messaging for performance-demand Grid Computing systems. For a moment, we have many Web Service tools such as XML, SOAP, SAX, DOM, XPATH, XSLT, etc, which are very high-level and can be improved for developing low latency high-performance Grid-communications. For example, the canonicalization technique of the original XML documents can speed up the time to process XML documents by a factor of two. Unfortunately, so far, there is no standard XSL output method which could be used with the identity transformer to generate a canonical form of a source document. Thus, this canonicalization can be researched and accomplished by my pre-processor experience and expertise of the HPspmd programming models. The other idea for speed-up of Grid Computing systems is that we need some different approaches for messaging, compared with legacy techniques. For instance, A/V sessions require some tricky set-up processes where the clients interested in participating, joining, and negotiating the session details. This part of the process has no significant performance issues and can be implemented with XML. The actual audio and video traffic does have performance-demands and here we can use existing faster protocol. Many applications consist of many control messages, which can be implemented in basic Web Service way and just the part of the messaging needs good performance. Thus, we end up with control ports running the basic WSDL with possible high-performance ports bound to a different protocol. This well-organized design and well-isolated messaging from the performance-demand applications can lead Grid to low latency high-performance Grid-communication environments. We call this kind of implementation and design issues to improve Grid Computing environments, the Web Service Compilation, or Grid-Compilation. We believe that Grid-Compilation will play the most important role in the entire Grid Computing environments since the performance-demands on heterogeneous systems are the most key issues for computational grids.
next up previous contents
Next: Current Status of HPJava Up: Future Works Previous: Java Numeric Working Group   Contents
Bryan Carpenter 2004-06-09