Grid Computing Research Laboratory State University of New York (SUNY) Binghamton Department of Computer Science

Tharaka Devadithya, Zongde Liu, Nayef Abu-Ghazaleh, Wei Lu, Kenneth Chiu, and Stephane Ethier
"BXSA for Fast Processing of Scientific Data",
Proceedings of the Seventh IEEE International Symposium on Cluster Computing and the Grid (CCGrid 2007),
Norfolk, Virginia, March 25-29, 2007
[PDF] [bibtex]

Abstract
XML has become the de facto standard for semi-structured data across a variety of domains. XML is generalliy considered to be slow for scientidic data and therefore only used for control information. However, this approach prevents scientific data from being "first class members" in XML, especially in a web services framework. It also puts a burden on application developers as they have learn two type systems. XML, however, can be represented via more efficient encodings, often known as "binary XML", which efficiently handles large data in XML format. In this paper we extend our previous work on Binary XML for Scientific Applications (BXSA) (1) applying BXSA to the Gyrokinetic Toroidal Code fusion application, and showing that performance is better than HDF5 in our test cases; (2) demonstrating an interoperable Java implementation that is faster than Xerces and Sun!/s Fast Infoset on common document types; and (3) showing that BXSA is also applicable to business data in addiotion to scientific data by evaluating its performance on a variety of XML documents against libxml2 and expat. These results show that BXSA is suitable for both business and scientific data.

Key Words
binary xml, xml, scientific computing, web services, high performance computing