CS 3520
Advanced Topics in Computer Networks: Performance
Analysis
Fall 2006
Syllabus
Instructor: Prof. José Carlos Brustoloni (jcb@cs.pitt.edu)
-- Classes: SENSQ 6516 – M W
-- Office hours: SENSQ
6111 – M W
Course rationale and objectives:
New types
of computer systems and networks usually are adopted only if they perform
better than previous alternatives, or offer other advantages (such as better
security or ease-of use) without excessive performance penalty. Performance analysis is therefore a necessary
step in many areas of computer science research. This course covers
systematically the main techniques used in performance analysis, including
workload selection and characterization, experimental design and analysis,
simulation, and queueing models.
Pre-requisites:
CS 1550 (undergraduate operating systems course) or
CS 1652 (undergraduate networking course)
Topics:
1. Introduction
a. Common mistakes
b. Selection of techniques and metrics
2. Workload selection and characterization
3. Monitors and accounting logs
4. Capacity planning and benchmarking
5. Data presentation and ratio games
6. Summarizing measured data
7. Comparing systems using sample data
8. Linear and other regression models
9. Experimental design
a. 2k factorial
b. 2kr factorial with replications
c. 2k-p fractional factorial
d. One-factor
e. Two-factor full factorial without replications
f. Two-factor full factorial with replications
g. General full factorial with k factors
10. Simulation
a. Basic tools
b. Model verification and validation
c. Transient removal and termination criteria
d. Variance reduction
e. Random-number generation
f. Random-variate generation
g. Commonly used distributions
11. Queueing models
a. Basic notation and rules; Little’s law
b. Analysis of a single queue
c. Queueing networks
d. Operational laws
e. Mean-Value Analysis (MVA)
f. Convolution algorithm
g. Hierarchical decomposition of large queueing networks
Required textbook:
Raj Jain, “The Art of Computer Systems Performance Analysis,” Wiley, available from the Pitt bookstore.
Grading:
30% Class presentations and participation
30% Homeworks
20% Midterm exam (11/13, covers chapters 1-18)
20% Final exam (12/13, covers chapters 19-34)
In some cases, it may be possible to develop a project
of appropriate scope in lieu of the exams. These cases need to be discussed
with and formally approved by the instructor before
Important dates:
8/28 First class
9/4 Labor Day (no class)
9/8 Add/Drop deadline
10/27 Withdrawal deadline
11/22-11/26 Thanksgiving recess (no class)
12/13 Last class
Policies:
·
Students should check regularly for course
announcements on http://courseweb.pitt.edu
·
Class attendance is necessary for class
participation. That portion of a student’s grade will suffer if the student
does not come to class.
·
Presentation slides should be emailed to the
instructor by
· Homeworks should be handed in to the instructor, printed or handwritten, by the beginning of class on the day it’s due (no email or mailbox, please). Late homeworks will be penalized 5% per day or fraction thereof up to two days and 10% per day or fraction up to another 5 days. They are unacceptable after that.
· Students caught cheating will fail the course.
· Except in case of documented emergency, there will be no make-up exams. Related documentation must be produced to the instructor no later than a week after a missed date.
Religious observances:
If a
scheduled class or exam conflicts with a religious observance you have, please
alert the instructor as early as possible in the term for rescheduling or other
accommodation.
Students with disabilities:
If you have a disability for which you are or may
be requesting an accommodation, you are encouraged to contact both your
instructor and the Office of Disability Resources and Services, 216 William
Pitt