Παρουσίαση/Προβολή
Performance evaluation and simulation
(ICSD104) - Prof Asimakis Leros
Περιγραφή Μαθήματος
Για να ξεκινήσουμε χωρίς παρανοήσεις, θα πρέπει να αναφέρω δύο ακόμη πράγματα. Ο φόρτος του μαθήματος είναι ομοιόμορφα μοιρασμένος από μέσα Οκτωβρίου μέχρι τέλη Ιανουαρίου. Στη διάρκεια αυτή θα έχετε να εκπονήσετε περίπου 8 εργασίες. Κάθε εργασία ξεκινάμε να την δουλεύουμε μαζί στο μάθημα, και την επόμενη εβδομάδα φέρνετε τα αποτελέσματά σας, τα οποία τα συζητάμε επίσης στο μάθημα. Αυτό σημαίνει ότι θα πρέπει να είστε παρόντες και παρούσες τις περίπου 12 Πέμπτες και 3--4 Τετάρτες που θα βρεθούμε. Αν λείψετε μία φορά (εσείς και τα αποτελέσματά σας), παρακαλώ ενημερώστε νωρίτερα.
Η δεύτερη παρατήρηση είναι ότι στις εργασίες δέν υπάρχουν σωστές και λάθος λύσεις, αλλά καλές και κακές λύσεις. Καλή είναι μια λύση που μπορεί να είναι λαθεμένη, αλλά την έχετε δουλέψει μόνοι σας και είστε σε θέση να την υποστηρίξετε και να την εξηγήσετε σε ένα τρίτο (δηλ. εμένα). Κακή είναι μια λύση (πιθανόν ολόσωστη) που την έχετε πάρει έτοιμη και δεν έχετε καθήσει αρκετές ώρες να την κάνετε τρείς φορές καλύτερη.
Αν τα παραπάνω δεν σας ταιριάζουν, μή δυσκολεύετε τη ζωή σας. Αφήστε το μάθημα για του χρόνου ή επιλέξτε κάποιο άλλο που σας ταιριάζει περισσότερο. Ακολουθεί η περιγραφή του μαθήματος.
Our subject in this course will be the common theory that can be used to describe several diverse systems, such as
- communication networks
- computing systems
- transportation networks
- production lines
- logistics / supply chain management
- customers served at facilities such as banks, post offices, super markets, gas stations etc.
The above applications are collectively termed "discrete event dynamical systems". They are characterized by discrete events that occur at random time instants, such as arrival and departure of packages at a communication node, arrival and departure of vehicles at a crossroad, equipment failure at a manufacturing facility etc.
Our goal is, based on some known parameters such as the rate of packet arrivals at a communications server, to compute performance measures such as mean delay time, blocking probability, throughput etc. There are two general methodologies to achieve this:
- Development of mathematical / statistical models of our system, and use of mathematical tools to compute the performance measures of interest. These tools are Markov processes and queueing theory; we will see them during the first few lectures.
- Simulation, i.e. implementation of a model of our system in software and execution of the software in order to reach some conclusions on the system performance. We will be using Matlab and Arena for our simulations.
The emphasis of this course is not on analyzing communication networks, but rather on discussing general principles. Some questions that we will address are:
- A server is idle for 50% of its time, but still queues keep forming. How can we explain this?
- Which is more efficient, a fast server or two slow ones?
- Which is the most efficient way to split incoming traffic into two servers?
- How do priorities affect performance?
See below for more details.
Ημερομηνία δημιουργίας
Τρίτη 1 Μαρτίου 2011
-
Course content
The course covers roughly the following topics (not necessarily in that order):
- Discrete event simulation. Structure of a simulation program.
- Poisson arrival processes. Birth and death processes. Markov processes. Equilibrium equations, steady state probabilities. Simulation of a Markov process.
- Single queues: M/M/1, M/M/1/K, M/M/c, M/M/1/K/K.
- Simulation of single queues using Matlab.
- The Arena simulation environment.
- General results for networks of queues. Product form networks.
- Simulation of networks of queues.
- Statistical analysis of simulation results. Confidence intervals. Hypothesis testing.
Course details typically vary from year to year.
Reference books
- A.O. Allen, Probability, Statistics and Queueing Theory with Computer Science Applications. Academic Press, 1990. (10 copies at the library)
- W.D. Kelton et al, Simulation with Arena, McGraw Hill 2015. (3 copies at the library)
- Θ. Δημητράκος, Πιθανότητες Στατιστική και Στοχαστικά Μοντέλα. Εκδόσεις Τσότρας 2025 (διαθέσιμο μέσω του Εύδοξου) --> covers probability, stochastic processes and statistics, same material as Chapter 3, 4, and 7 of Allen.
- Σταφυλοπάτης & Σιόλας, Ανάλυση Επίδοσης Υπολογιστικών Συστημάτων. (ηλεκτρονικό, διαθέσιμο μέσω του Κάλλιππου) --> queueing theory, same as Chapters 5 and 6 of Allen.
- Κουικόγλου & Κωνσταντάς, Προσομοίωση συστημάτων διακριτών γεγονότων. Εκδόσεις Δίσιγμα 2016 (διαθέσιμο στη βιβλιοθήκη) --> has an introduction to Arena, but much less detailed than Kelton.
- Harry Perros, Computer Simulation Techniques: The definitive introduction. (available at http://www.csc.ncsu.edu/faculty/perros//simulation.pdf) --> covers general simulation techniques (not Arena).
We will be using the books by Allen and Kelton as reference, not textbooks.
Course grade
Seven or eight lab assignments (50%)
Assignments will be mostly programming with some theoretical questions. The emphasis is not on implementation, but rather on producing results and reaching conclusions. You will need a basic knowledge of Matlab for the first assignments. It goes without saying that you may contact us if you run into difficulties.
Assignment grades (A/B/C/D/F) are based on all of the following:
- completeness of lab report
- implementation
- presentation of results
- comments and conclusions
- active participation during the lab
Mid-term exam (20%)
A 2 hour paper exam, with no access to books, notes or computing equipment, but with an extensive formula sheet. Will fall on a Wednesday, likely Nov 26.
Programming (simulation) project (30%)
You will start by the end of November and present your results by the end of January. Choose a subject from the course literature, something interesting you may find on the Web (even if it has already been implemented), or any other source.
Remark 1: You may discuss assignments, but each one has to present his/her own work. Lab reports with similarities beyond statistical coincidence will not be graded.
Remark 2: I must have checked your implementation and results before you submit a lab report.
Prerequisites
- Probability and random variables
- Programming in Matlab
Instructor
Prof Asimakis Leros
ICSD Building, office B10
aleros at aegean dot gr
Office hours: Wed 18:00--20:00, Thu 13:00--15:00