Probabilistic Scheduling Based on Queueing Model for Multi-user Network Applications

Multi-user network applications are often met in our normal life, especially as WEB comes up. And it is usually configured into a distributed system. In such distributed environment, task scheduling plays an important role. Unlike traditional scheduling, there are lots of uncertainties which consist in task arrival process, state of processing units, and network transmission. We investigate the dynamic scheduling problem which takes these uncertainties into account. In order to improve scheduling speed and alleviate the problem of imprecision of state estimation, the idea of probabilistic scheduling is put forward. Specifically, we build a queueing model in which all arrived tasks are pushed into a queue, and every processing unit has their local queues for task execution. Some parameters are computed using queueing theory. Based on these parameters this scheduling problem is defined as a non-linear programming problem. The robustness of the algorithm is analyzed, and the performance is validated by comparison with such classical algorithms as MinMin, MinMax, Suffrage, and ECT.