Abstract
In this thesis, we present two important new results for system softwares of parallel computations. These two results are done in my first year and second year MS work, respectively.In the first work, we propose workload models to schedule real-time works with gang scheduling environments. Gang scheduling has recently been shown to be an effective jobscheduling policy for parallel computers that combines elements of space sharing and time sharing\cite{FeRu92,Setia97}.In this work, we propose new policies to enable gang scheduling to adapt to environments with real-time constraints.Our work, to our best knowledge, is the first work to attempt to address the real-time aspects with gang scheduling. Our system guided by a metric, called ``task utilization workload'', can schedule both real-time and non-real-time tasks at the same time. In our work, we report simulation results with a family of scheduling algorithms based on our proposed metric. Our scheme is designed to be a practical scheme to be used for large scale industrial and commercial parallel systems. Preliminary simulation results also show that our proposed policy is an effective scheme to perform real-time scheduling, while schedule non-real-time jobs with fairness and good throughput.This work is jointly done with Prof. Wei-Kuan Shih's Group during an industrial research project.In the second work, we implement a pointer analysis for distributed data structures. For many parallel scientific computations, the performance of the program ca be improved, if compiler can obtain the data structure information from user programs. The theoretical aspect of this work was published earlier in ICPP'97 by our Lab members and received the mostoriginal paper award in ICPP'97. In this work, we focus onthe implementation of the pointer analysis to verify the theories with actual experimental results. We incorporate PIP (Parametric Integer Programming) solver to analyze the data distribution of pointer-based programs. With our implementation,we now shoe that the theory developed earlier by out Lab members is mapping data distribution analysis problem into PIP(Parametric Integer Programming) solver can be implemented effectively and efficiently.