Grid Computing Research Laboratory

State University of New York (SUNY) Binghamton
Department of Computer Science

[Home] [Background] [For Visitors] [Projects] [People] [Papers] [Talks] [Resources] [Funding] [Internal]

Tharaka Devadithya and Kenneth Chiu
"Fast Binary Serialization for Grid Systems with XBS",
SC06 (Supercomputing): International Conference for High Performance Computing, Networking and Storage,
Tampa, Florida, November 11-17, 2006.
[PDF] [bibtex]

Abstract
Efficient serialization and deserialization of data is a fundamental operation of many grid systems. Some serializers are message-based, while others are stream-based. Streaming serializers can be more scalable and flexible, by promoting free form conversations not fixed to any particular static structure. We present the design and implementation of the XBS binary serializer, focusing on three important features: efficient serialization of large and small arrays, efficient pass-through of opaque data by grid intermediaries, and efficient representation of numbers with a large dynamic range. The first feature is important because such arrays dominate scientific computing. The second feature is useful for grid intermediaries such as gateways and proxies, which are becoming ever more important as grid systems become more complex. The third feature is important for efficiently supporting systems without arbitrary size limits. XBS is a freely-available C++ library and provides an object-oriented API, based on generic programming techniques.

Key Words