System Support to Balance the Resource Supply and Demand in High-end Computing

Authors:
Xiaodong Zhang
Affiliations:
College of William and Mary, Williamsburg, VA
Venue:
IPDPS '05 Proceedings of the 19th IEEE International Parallel and Distributed Processing Symposium (IPDPS'05) - Workshop 10 - Volume 11
Year:
2005

Citing 8
Cited 0

LogP: towards a realistic model of parallel computation

PPOPP '93 Proceedings of the fourth ACM SIGPLAN symposium on Principles and practice of parallel programming
Latency metric: an experimental method for measuring and evaluating parallel program and architecture scalability

Journal of Parallel and Distributed Computing - Special issue on scalability of parallel algorithms and architectures
Memory Hierarchy Considerations for Cost-Effective Cluster Computing

IEEE Transactions on Computers
LIRS: an efficient low inter-reference recency set replacement policy to improve buffer cache performance

SIGMETRICS '02 Proceedings of the 2002 ACM SIGMETRICS international conference on Measurement and modeling of computer systems
Dynamic Cluster Resource Allocations for Jobs with Known and Unknown Memory Demands

IEEE Transactions on Parallel and Distributed Systems
Adaptive Memory Allocations in Clusters to Handle Unexpectedly Large Data-Intensive Jobs

IEEE Transactions on Parallel and Distributed Systems
Latency lags bandwith

Communications of the ACM - Voting systems
CLOCK-Pro: an effective improvement of the CLOCK replacement

ATEC '05 Proceedings of the annual conference on USENIX Annual Technical Conference

Quantified Score

Hi-index	0.00

Visualization

Abstract

Modern and large scale high-end systems built by commercial processors can easily provide unbalanced computing resources with over-supplied CPU cycles and increasingly long latency of data accesses at different levels of the memory hierarchy for high demands of data-intensive applications. The imbalance is mainly caused by two technology changes. First, the speed gap between the CPU and the memory and the I/O storage continues to grow. Second, the latency improvement significantly lags behind the bandwidth improvement in any pair connection of the computer systems. We are working on three research projects to cope with the imbalance problem. (1) We develop models to characterize the increasingly complex and deep memory hierarchy, providing critical insights and guidance for high-end systems design. (2) We develop several memory-centric load-sharing schemes and their implementations for high end-systems management. (3) We develop efficient locality-aware caching algorithms and their system implementations for virtual memory and buffer cache management. This paper briefly report the background, motivation, and working progress of our research funded by the NSF NGS program.