Queueing Disciplines

The goal of scheduling is to reduce mean delay for free! By simply serving jobs in the “right order,” we can often reduce mean delay by an order of magnitude, without requiring the purchase of additional resources. How can this be? What is the “right order” in which jobs should be scheduled? The answer depends on parameters of the system being scheduled, like load and the job size distribution. These parameters and others will be explained in Section 2. Typical scheduling policies will then be defined in Section 3 and analyzed in Section 4. Finally, in Section 5 we will see some recent real-world success stories where scheduling has been used to greatly decrease average delays. The versatility of scheduling has made it the topic of thousands of research papers. While the majority of this paper introduces scheduling under the simplest queueing model (an M/G/1 queue) and considers only the simplest performance metric (mean response time), in Section 6 we briefly discuss more complex queueing models, performance metrics, and scheduling policies, and give pointers to research papers covering those topics.