Placement of Processes and Files in Distributed Systems (thesis)
In this thesis, we examine ways of improving the average job performance in a distributed system by controlling on which machine processes or files are placed. We describe an implementation of process migration under Sun UNIX, and use this implementation in experiments to assess the relative merits of process migration and initial placement as load-balancing strategies. Finally, we examine dynamic file caching in distributed systems where maintaining file consistency is important. Using an analytical model, we derive a general caching algorithm. In a series of simulation studies, we compare the performance
of this algorithm with that of the simpler strategies of never caching and always caching files for various network configurations.