next up previous contents
Next: Scalable and high-performance file Up: RESEARCH ACTIVITIES Previous: A MATLAB Compilation Environment

Runtime Systems and Optimizations for High-performance I/O (A. Choudhary - PASSION Project)

In the MiPFS project, on Multimedia Integrated Parallel File System we have designed a parallel file system for network of workstations. The main features of this design are to support basic file system operations for scientific and multimedia applications in an integrated manner. The design supports quality of service requirements for requests as well as best effort for scientific applications. It supports hierarchical parallelism for accessing striped files.

In the BESS work, we introduce the bit-encoded sparse structure (BESS) for storing multi-dimensional sparse data. We developed an analysis of some typical OLAP operations using BESS and other sparse storage schemes for higher dimensional data. Performance results are presented, clearly showing the advantages of BESS over other sparse data structures. We also evaluate BESS with a space times time metric and find that access performance is better than multi-dimensional arrays for data densities up to 15, usually found for high dimensional data in practice.

This research has been supported by Intel, the DOE, DARPA, and NSF and has supported 4 Ph.D. students (one female), and 1 M.S. student (minority). One Ph.D. (Meena Kandaswamy) and one M.S. student (Juan Suaris) has graduated.


next up previous contents
Next: Scalable and high-performance file Up: RESEARCH ACTIVITIES Previous: A MATLAB Compilation Environment
Tom Conversion Service