Date Awarded

2002

Document Type

Dissertation

Degree Name

Doctor of Philosophy (Ph.D.)

Department

Computer Science

Advisor

Xiaodong Zhang

Abstract

The effectiveness of computer system resource management has been always determined by two major factors: (1) workload demands and management objectives, (2) the updates of the computer technology. These two factors are dynamically changing, and resource management systems must be timely adaptive to the changes. This dissertation attempts to address several important and related resource management issues.;We first study memory system utilization in centralized servers by improving memory performance of sorting algorithms, which provides fundamental understanding on memory system organizations and its performance optimizations for data-intensive workloads. to reduce different types of cache misses, we restructure the mergesort and quicksort algorithms by integrating tiling, padding, and buffering techniques and by repartitioning the data set. Our study shows substantial performance improvements from our new methods.;We have further extended the work to improve load sharing for utilizing global memory resources in distributed systems. Aiming at reducing the memory resource contention caused by page faults and I/O activities, we have developed and examined load sharing policies by considering effective usage of global memory in addition to CPU load balancing in both homogeneous and heterogeneous clusters.;Extending our research from clusters to Internet systems, we have further investigated memory and storage utilizations in Web caching systems. We have proposed several novel management schemes to restructure and decentralize the existing caching system by exploiting data locality at different levels of the global memory hierarchy and by effectively sharing data objects among the clients and their proxy caches.;Data integrity and communication anonymity issues are raised from our decentralized Web caching system design, which are also security concerns for general peer-to-peer systems. We propose an integrity protocol to ensure data integrity, and several protocols to achieve mutual communication anonymity between an information requester and a provider.;The potential impact and contributions of this dissertation are briefly stated as follows: (1) two major research topics identified in this dissertation are fundamentally important for the growth and development of information technology, and will continue to be demanding topics for a long term. (2) Our proposed cache-effective sorting methods bridge a serious gap between analytical complexity of algorithms and their execution complexity in practice due to the increasingly deep memory hierarchy in computer systems. This approach can also be used to improve memory performance at different levels of the memory hierarchy, such as I/O and file systems. (3) Our load sharing principle of giving a high priority to the requests of data accesses in memory and I/Os timely adapts the technology changes and effectively responds to the increasing demand of data-intensive applications. (4) Our proposed decentralized Web caching framework and its resource management schemes present a comprehensive case study to examine the P2P model. Our results and experiences can be used for related and further studies in distributed computing. (5) The proposed data integrity and communication anonymity protocols address limits and weaknesses of existing ones, and place a solid foundation for us to continue our work in this important area.

DOI

https://dx.doi.org/doi:10.21220/s2-deqc-ew25

Rights

© The Author

Share

COinS