Computer Science Department
School of Computer Science, Carnegie Mellon University
Affinity Scheduling in Staged Server Architectures
Stavros Harizopoulos, Anastassia Ailamaki
This technical report introduces (a) four novel cohort scheduling techniques for staged software servers that follow a "production-line" model of operation, and (b) a mathematical framework to methodically quantify the performance trade-offs when using these techniques. Our Markov chain analysis of one of the scheduling techniques matches the simulation results. Using our model on a staged database server, we found that the proposed policies exploit data and instruction locality for a wide range of workload parameter values and outperform traditional techniques such as FCFS and processor-sharing. Consequently, our results justify the restructuring of a wide class of software servers to incorporate the staged programming paradigm.