Enabling Circuit Switching
PIs: Alex K. Jones and Rami Melhem
Advances in optical devices and interconnects have made this technology attractive for telecommunication and high performance systems. Specifically, large optical switches are commercially available but no practical technology is yet available for buffering or processing in the optical domain. As a result, electronic packet switching and wormhole-routing remain popular in spite of their higher costs, limited bandwidth, and increased message latencies.
In this project, the Investigators propose the use of circuit switching as an efficient alternative to packet and wormhole switching for achieving high bandwidth, low latency communication in high performance parallel systems. Specifically, by establishing direct pipes between communicating processes, routing and buffering at intermediate switches can be eliminated, end-to-end protocols can be simplified, and both software and hardware overheads associated with data movement can be minimized. However, the overhead of circuit establishment can be relatively large, and the benefits of circuit switching can only outweigh its drawbacks when communication exhibits locality, and when this locality is appropriately explored. Thus, a compiler-based approach is proposed to analyze communication locality in parallel application programs. This approach exposes the communication locality to an architecture, which explores it through a run-time system that manages circuit switched interconnections.
The new compiler technology for the detection and manipulation of communication patterns developed as part of this research enables parallelism in several domains outside of high performance supercomputers including chip multiprocessor systems and wireless systems. By correlating code structure with communication patterns, the proposed compiler technology also allows for the construction of tools that assist application developers in writing more efficient programs. Moreover, this project impacts several different areas of curriculum in computer science and engineering through the integration of research from this project into computer engineering curriculum and design projects. The project also promotes summer research activities related aimed at active recruitment and participation of women and under-represented minorities in higher education and research.
This project is supported by the National Science Foundation.
Related Publications
•
•
•R. Hoare, Z. Ding, and A. K. Jones, “A Near-optimal Two-stage Hardware Scheduler for Large Cardinality Crossbar Switches,” Journal of Parallel and Distributed Computing (JPDC) - in press.
•S. Shao, A. K. Jones, and R. Melhem, “Compiler Techniques for Efficient Communications in Circuit Switch Networks for Multiprocessor Systems,” IEEE Transactions on Parallel and Distributed Systems, pre-print. [pdf]
•R. Melhem, “Low Diameter Interconnections for Routing in High Performance Parallel Systems,” IEEE Transactions on Computers, Vol. 56, No. 4, April 2007, pp. 502-510. [pdf]
•R. Hoare, Z. Ding, S. Tung, Rami Melhem, and A. K. Jones, “A Framework for the Design, Synthesis and Cycle-Accurate Simulation of Multiprocessor Networks,” Journal of Parallel and Distributed Computing, Vol. 65, No. 10, October 2005, pp. 1237-1252. [pdf]
•S. Shao, Y. Zhang, A. K. Jones, and R. Melhem, “Symbolic Expression Analysis for Compiled Communication,” in Proc. of the IPDPS Workshop on Large Scale Parallel Processing (LSPP), April 2008. [pdf]
•R. Hoare, Z. Ding, A. K. Jones, “A Near-optimal Real-time Hardware Scheduler for Large Cardinality Crossbar Switches,” in Proc. of Supercomputing (SC), 2006. [pdf]
•S. Shao, A. K. Jones, R. Melhem, “A Compiler-based Communication Analysis Approach for Multiprocessor Systems,” in Proc. of IEEE/ACM International Parallel and Distributed Processing Symposium (IPDPS), 2006. [pdf]
•K. J. Barker, A. Benner, R. Hoare, A. Hoisie, A. K. Jones, D. J. Kerbyson, D. Li, R. Melhem, R. Rajamony, E. Schenfeld, S. Shao, C. Stunkel, and P. A. Walker, “On the Feasibility of Optical Circuit Switching for High Performance Computing Systems,” IEEE/ACM Supercomputing Conference (SC), 2005. [pdf]
•R. Melhem, R. Hoare, A. Jones, Z. Ding, S. Tung, D. Li, S. Shao, J. Zheng, “Switch Design to Enable Predictive Multiplexed Switching in Multiprocessor Networks,” in Proc. of the International Parallel & Distributed Processing Symposium (IPDPS), 2005. [pdf]