Task Scheduling in Grid Computing Systems constrained by Resource Availability

In this paper, a survey of some of the approaches to task scheduling in grid computing systems based on resource availability is discussed. To improve task scheduling, resource availability can be predicted by multistate availability predictors. Execution time of each application is modeled by a random variable. Each application is associated with a set of probability mass function (pmf), one pmf for each machine in the Heterogeneous computing suite, describing the probability of all possible execution times for that application. Stochastic completion times require the summation of random variables. The MaxRobust heuristic applies a simple greedy approach to maximizing the stochastic robustness of the resource allocation. A 3 state Markovian model is used for scheduling iterative master worker applications on processors that can fail or be temporarily reclaimed. Resource selection involves several factors such as resource access policies, status of execution, and computation capability of resource provider. The two main approaches in resource selection: centralized matchmaking-based approach and local knowledge-based approach are discussed.