A process migration mechanism offers a means to exploit the performance reserves present in networks of workstations used as personal computers by allowing the processors to migrate processes from overload ones to underused ones. Several distributed operating systems provide such a facility, the benefits of its use depending on the specification of a proper process migration policy. This work proposes an analytical model, the Markov Team Model, to assist the design of such a policy. Besides deriving this model from results of classical Team Theory and Markov Decision Processes, we study the special case of homogeneous distributed computing systems and present methods for parameter estimation. Numerical examples are used to demonstrate the benefits of using this model.
Process Migration in Distributed Computing SystemsA human team is characterized as a collection of individuals with a common goal that cooperate on a particular task. The processors of a distributed computing system can be programmed to behave as a team, too, by implementing a mechanism that allows them to share the work load. Such a process migration facility is particularly attractive when considering a network of workstations used as personal computers, where typically a large fraction of the processors are unused or underused at any instance of time. Thus, a number of distributed operating systems 2 contain software that allows the migration of processes between processors. Typically, such a process migration mechanism is implemented in the kernel of a distributed operating system, whereby the design goals of a process migration mechanism as posed to the implementor are transparency to the user, overhead minimization and fault tolerance (for the discussion of these goals and the design alternatives available see e.g. Douglis, Ousterhout (1991),