Monte Carlo Methods and Appl., Vol. 11, No.1, pp. 39 – 55 (2005) c  VSP 2005 Grid-based Quasi-Monte Carlo Applications Yaohang Li 1 and Michael Mascagni 2 1 Department of Computer Science, North Carolina A&T State University, Greensboro, NC 27411 USA, yaohang@ncat.edu 2 Department of Computer Science and School of Computational Science, Florida State University, Tallahassee, FL 32306-4560 USA, mascagni@fsu.edu Abstracts — In this paper, we extend the techniques used in Grid-based Monte Carlo appli- cations to Grid-based quasi-Monte Carlo applications. These techniques include an N-out-of-M strategy for efficiently scheduling subtasks on the Grid, lightweight checkpointing for Grid sub- task status recovery, a partial result validation scheme to verify the correctness of each individual partial result, and an intermediate result checking scheme to enforce the faithful execution of each subtask. Our analysis shows that the extremely high uniformity seen in quasirandom sequences prevents us from applying many of our Grid-based Monte Carlo techniques to Grid- based quasi-Monte Carlo applications. However, the use of scrambled quasirandom sequence becomes a key to tackling this problem, and makes many of the techniques we used in Grid- based Monte Carlo applications effective in Grid-based quasi-Monte Carlo applications. All the techniques we will describe here contribute to performance improvement and trustworthiness enhancement for large-scale quasi-Monte Carlo applications on the Grid, which eventually lead to a high-performance Grid-computing infrastructure that is capable of providing trustworthy quasi-Monte Carlo computation services. 1. Introduction Grid computing is characterized by large-scale sharing and cooperation of dynamically distributed resources, such as CPU cycles, communication bandwidth, and data, to con- stitute a computational environment [12]. In the Grid’s dynamic environment, from the application point-of-view, two issues are of prime import: performance – how quickly the Grid-computing system can complete the submitted tasks, and trustworthiness – that the results obtained are, in fact, due to the computation requested. To meet these two re- quirements, many Grid-computing or distributed-computing systems, such as Condor [17], HARNESS [6], Javelin [20], Globus [11], and Entropia [2], concentrate on developing high- performance and trust-computing facilities through system-level approaches. In [14], we analyzed the characteristics of Monte Carlo applications to develop approaches to address