CMSC 710 PERFORMANCE EVALUATION OF COMPUTER SYSTEMS
Catalog Description
A comprehensive study of modelling distributed systems and evaluating their
performance. Topics include stochastic processes, queueing theory,
discrete-event simulation, synthetic workload generation, measurement and
analysis, benchmarking programs.
Prerequisites
CMSC 411 and CMSC 412 or equivalent. Calculus and
probability theory recommended.
Topics
Elementary statistics: basic sampling methods, clustering, regression analysis, stratified sampling, sensitivity analysis and order statistics.
Benchmarking, Simulation and Analysis: Trade-offs Between Abstraction and Coverage.
Experimental Design: Single and Multiple-Factor Experiments, Full-Factorial Designs, 2kr-Factorial Designs, Effects and Confidence Intervals.
Stochastic processes and queueing theory: discrete and continuous-time queues; workload models and queue-selection policies.
Modelling and benchmarking methods, as described in case-study papers.
Course Text
The Art of Computer Systems Performance Analysis Techniques for
Experimental Design, Measurement, Simulation, and Modeling. Jain,
John Wiley & Sons, 1991, ISBN 0471503363.
Workload and Grading
Group project [50%]:
The course incorporates a hands-on performance
study, which will be carried out in group projects.
Each group will select the specific system to be analyzed, in
consultation with the professor. Several sample systems will be
suggested, but the exact choice will be made by each group.
Individual projects will be permitted in
special cases;
these will have a more significant analytical emphasis.
Evaluations will be based on the soundness of the experimental design and
the methods used to carry it out -- not necessarily on
the level of success achieved.
Review of a case study paper [15%]: The paper
topic should help provide background for your selected project.
Midterm [15%] and Final [20%]: These will contain questions
on the core theoretical material; the
case study papers reviewed in class; and some questions on
your specific group project, to
help gauge your individual participation.
Sample homework problems will also be assigned periodically; however,
these will not be graded.
