Basic Queueing Theory M/M/* Queues. Introduction



Similar documents
1. Fundamentals of probability theory 2. Emergence of communication traffic 3. Stochastic & Markovian Processes (SP & MP)

BANDWIDTH ALLOCATION AND PRICING PROBLEM FOR A DUOPOLY MARKET

Stochastic Models of Load Balancing and Scheduling in Cloud Computing Clusters

An Analysis of Central Processor Scheduling in Multiprogrammed Computer Systems

Stochastic Models of Load Balancing and Scheduling in Cloud Computing Clusters

An Electricity Trade Model for Microgrid Communities in Smart Grid

Inventory Control in a Multi-Supplier System

Stochastic Models of Load Balancing and Scheduling in Cloud Computing Clusters

A Statistical Model for Detecting Abnormality in Static-Priority Scheduling Networks with Differentiated Services

A R T I C L E S DYNAMIC VEHICLE DISPATCHING: OPTIMAL HEAVY TRAFFIC PERFORMANCE AND PRACTICAL INSIGHTS

Performance Analysis of Energy Consumption of Smartphone Running Mobile Hotspot Application

Recurrence. 1 Definitions and main statements

How Much to Bet on Video Poker

An Analytical Model of Web Server Load Distribution by Applying a Minimum Entropy Strategy

Quality of Service Analysis and Control for Wireless Sensor Networks

Description of the Force Method Procedure. Indeterminate Analysis Force Method 1. Force Method con t. Force Method con t

benefit is 2, paid if the policyholder dies within the year, and probability of death within the year is ).

PSYCHOLOGICAL RESEARCH (PYC 304-C) Lecture 12

This paper concerns the evaluation and analysis of order

Power-of-Two Policies for Single- Warehouse Multi-Retailer Inventory Systems with Order Frequency Discounts

Ganesh Subramaniam. American Solutions Inc., 100 Commerce Dr Suite # 103, Newark, DE 19713, USA

Retailers must constantly strive for excellence in operations; extremely narrow profit margins

Analysis of Energy-Conserving Access Protocols for Wireless Identification Networks

The Greedy Method. Introduction. 0/1 Knapsack Problem

Module 2 LOSSLESS IMAGE COMPRESSION SYSTEMS. Version 2 ECE IIT, Kharagpur

BERNSTEIN POLYNOMIALS

Causal, Explanatory Forecasting. Analysis. Regression Analysis. Simple Linear Regression. Which is Independent? Forecasting

TheHow and Why of Having a Successful Home Office

THE DISTRIBUTION OF LOAN PORTFOLIO VALUE * Oldrich Alfons Vasicek

What is Candidate Sampling

The OC Curve of Attribute Acceptance Plans

1 Example 1: Axis-aligned rectangles

CALL ADMISSION CONTROL IN WIRELESS MULTIMEDIA NETWORKS

INTRODUCTION TO MERGERS AND ACQUISITIONS: FIRM DIVERSIFICATION

Optimal outpatient appointment scheduling

Enabling P2P One-view Multi-party Video Conferencing

Conferencing protocols and Petri net analysis

v a 1 b 1 i, a 2 b 2 i,..., a n b n i.

Maximizing profit using recommender systems

AN APPOINTMENT ORDER OUTPATIENT SCHEDULING SYSTEM THAT IMPROVES OUTPATIENT EXPERIENCE

Rapid Estimation Method for Data Capacity and Spectrum Efficiency in Cellular Networks

Small-Signal Analysis of BJT Differential Pairs

Extending Probabilistic Dynamic Epistemic Logic

II. THE QUALITY AND REGULATION OF THE DISTRIBUTION COMPANIES I. INTRODUCTION

Packet Reorderng Analysis

MAC Layer Service Time Distribution of a Fixed Priority Real Time Scheduler over

Revenue Maximization Using Adaptive Resource Provisioning in Cloud Computing Environments

Formulating & Solving Integer Problems Chapter

Vasicek s Model of Distribution of Losses in a Large, Homogeneous Portfolio

Robust Design of Public Storage Warehouses. Yeming (Yale) Gong EMLYON Business School

n + d + q = 24 and.05n +.1d +.25q = 2 { n + d + q = 24 (3) n + 2d + 5q = 40 (2)

A NOTE ON THE PREDICTION AND TESTING OF SYSTEM RELIABILITY UNDER SHOCK MODELS C. Bouza, Departamento de Matemática Aplicada, Universidad de La Habana

Little s Law & Bottleneck Law

Support Vector Machines

Development of TIF for transaction cost allocation in deregulated power system

8.5 UNITARY AND HERMITIAN MATRICES. The conjugate transpose of a complex matrix A, denoted by A*, is given by

Multiple stage amplifiers

Online Algorithms for Uploading Deferrable Big Data to The Cloud

DEFINING %COMPLETE IN MICROSOFT PROJECT

PERRON FROBENIUS THEOREM

Institute of Informatics, Faculty of Business and Management, Brno University of Technology,Czech Republic


Title: A Queuing Network Model with Blocking: Analysis of Congested Patient Flows in Mental Health Systems

21 Vectors: The Cross Product & Torque

Case Study: Load Balancing

Modeling and Assessment Performance of OpenFlow-Based Network Control Plane

How Sets of Coherent Probabilities May Serve as Models for Degrees of Incoherence

A Fuzzy Optimization Framework for COTS Products Selection of Modular Software Systems

Risk-based Fatigue Estimate of Deep Water Risers -- Course Project for EM388F: Fracture Mechanics, Spring 2008

Stochastic epidemic models revisited: Analysis of some continuous performance measures

A Novel Dynamic Role-Based Access Control Scheme in User Hierarchy

"Research Note" APPLICATION OF CHARGE SIMULATION METHOD TO ELECTRIC FIELD CALCULATION IN THE POWER CABLES *

Efficient Striping Techniques for Variable Bit Rate Continuous Media File Servers æ

THE METHOD OF LEAST SQUARES THE METHOD OF LEAST SQUARES

Sketching Sampled Data Streams

Least Squares Fitting of Data

Logical Development Of Vogel s Approximation Method (LD-VAM): An Approach To Find Basic Feasible Solution Of Transportation Problem

Technical Report, SFB 475: Komplexitätsreduktion in Multivariaten Datenstrukturen, Universität Dortmund, No. 1998,04

RELIABILITY, RISK AND AVAILABILITY ANLYSIS OF A CONTAINER GANTRY CRANE ABSTRACT

Ring structure of splines on triangulations

A Design Method of High-availability and Low-optical-loss Optical Aggregation Network Architecture

Section 5.4 Annuities, Present Value, and Amortization

PAS: A Packet Accounting System to Limit the Effects of DoS & DDoS. Debish Fesehaye & Klara Naherstedt University of Illinois-Urbana Champaign

Analysis of Clock Synchronization Approaches for Residential Ethernet

A Lyapunov Optimization Approach to Repeated Stochastic Games

Using Series to Analyze Financial Situations: Present Value

Course outline. Financial Time Series Analysis. Overview. Data analysis. Predictive signal. Trading strategy

CloudMedia: When Cloud on Demand Meets Video on Demand

1.1 The University may award Higher Doctorate degrees as specified from time-to-time in UPR AS11 1.

Web Service-based Business Process Automation Using Matching Algorithms

A Distributed Algorithm for Least Constraining Slot Allocation in MPLS Optical TDM Networks

A Probabilistic Theory of Coherence

1 OPTIMIZATION ISSUES IN WEB

Capacity Planning for Virtualized Servers

Value Driven Load Balancing

Chapter 22 Heat Engines, Entropy, and the Second Law of Thermodynamics

PRIOR ROBUST OPTIMIZATION. Balasubramanian Sivan. A dissertation submitted in partial fulfillment of the requirements for the degree of

An Adaptive Cross-layer Bandwidth Scheduling Strategy for the Speed-Sensitive Strategy in Hierarchical Cellular Networks

1. Measuring association using correlation and regression

Luby s Alg. for Maximal Independent Sets using Pairwise Independence

Transcription:

Basc Queueng Theory M/M/* Queues These sldes are created by Dr. Yh Huang of George Mason Unversty. Students regstered n Dr. Huang's courses at GMU can ake a sngle achne-readable copy and prnt a sngle copy of each slde for ther own reference, so long as each slde contans the copyrght stateent, and GMU facltes are not used to produce paper copes. ersson for any other use, ether n achnereadable or prnted for, ust be obtaned fro the author n wrtng. CS 756 Introducton Queueng theory provdes a atheatcal bass for understandng and predctng the behavor of councaton networks. Basc Model Arrvals Queue Server Departures CS 756

Major paraeters: nterarrval-te dstrbuton servce-te dstrbuton nuber of servers queueng dscplne how custoers are taken fro the queue, for exaple, FCFS nuber of buffers, whch custoers use to wat for servce A coon notaton: A/B/, where s the nuber of servers and A and B are chosen fro M: Markov exponental dstrbuton D: Deternstc G: General arbtrary dstrbuton CS 756 3 M/M/ Queueng Systes Interarrval tes are exponentally dstrbuted, wth average arrval rate. Servce tes are exponentally dstrbuted, wth average servce rate. There s only one server. The buffer s assued to be nfnte. The queung dscplne s frst-coe-frstserve FCFS. CS 756 4

Syste State Due to the eoryless property of the exponental dstrbuton, the entre state of the syste, as far as the concern of probablstc analyss, can be suarzed by the nuber of custoers n the syste,. the past/hstory how we get here does not atter When a custoer arrves or departs, the syste oves to an adjacent state ether + or -. CS 756 5 State Transton Dagra 3 4 5 In equlbru, Let { syste n state } We have + + The rate of oveents n both drectons should be equal CS 756 6 3

4 CS 756 7 Equatons fro the state transton dagra: Solve What s? 3 k k CS 756 8 Snce we have That s,. Note that ust be less than, or else the syste s unstable. k k k k. k k

5 CS 756 9 Average Nuber of Custoers? k k k k k k N CS 756 Average delay per custoer te n queue plus servce te: Average watng te n queue: Average nuber of custoers n queue: N T W W N Q

Applcatons Consder 4 coputer users, each of whch produces n average 48 packets per second. For every custoer, the nterarrval tes of hs packets are exponentally dstrbuted. The lengths of packets are also exponentally dstrbuted, wth ean 5 bytes. CS 756 Scenaro Users share a T lne usng the standard T te-dvson ultplexng. Assue that each user s assocated wth an nfnte buffer that s, queue. In a T lne, t takes /8 seconds to delver or serve each byte. However, due to ther varable lengths, the delvery or servce tes of packets are stll exponentally dstrbuted. The average servce rate? CS 756 6

The syste can be consdered as 4 M/M/ queues: acket Arrvals Queue Server, /4th of the T lne Departs to the other end of the T lne We have 48 75% 64.75 N 3.75 T 64 48 4 4sec CS 756 3 Scenaro Users share a.544 Mbps lne through an I router. ackets fro 4 users The entre T as the server CS 756 4 7

The aggregated arrval rate s 4 48 5. The servce rate s 4 64 536. We have 48/ 64 75% T 536 5 384.4 sec Ths syste s 4 tes faster than TDM! CS 756 5 Dscusson Flaws n the analyss? Stll such a drastc dfference n results convncngly reveals the neffcency of TDM. Ths partly explans the oentu toward usng the Internet as the unversal nforaton nfrastructure. In general, allowng custoers to share a pool of resources s far ore effcent than allocatng a fxed porton to each custoers. CS 756 6 8

M/M/ Queueng Systes Arrvals Queue Departs Servers All servers are dentcal, wth servce rate CS 756 7 State Transton Dagra 3 + Balance equatons:,, for for > CS 756 8 9

CS 756 9 Soluton Where Notcng that we have > p for,! for,!., [ ]!! + CS 756 The probablty that an arrvng custoer has to wat n queue: Ths s known as the Erlang C forula.!!! Q

CS 756 Average nuber of watng custoers: + +!!! N Q Q CS 756 Average watng te n queue: Average te n the syste: Average nuber of custoers n the syste: N W Q Q + + + W T Q Q + + T N Q Q

M/M//K Queueng Systes Slar to M/M/, except that the queue has a fnte capacty of K slots. That s, there can be at ost K custoers n the syste. If a custoer arrves when the queue s full, he/she s dscarded leaves the syste and wll not return. CS 756 3 Analyss Notce ts slarty to M/M/, except that there are no states greater than K. We have Notcng that K- K, for K, for > K K + K we have CS 756 4

osson rocess Let rando varable N be a counter of the nuber of occurrences of a partcular type of events. Clearly, the value of N ncreases over te. Let N t be the value of N at te t. Moreover, f N, Nt s sad to be a countng process. The countng process Nt s sad to be a osson process havng rate f the nuber of events n any nterval of length t s osson dstrbuted wth ean t. That s, for all s, t t t { N t + s N s n} e, n,,... n! CS 756 5 n Dscusson The nterarrval tes of a osson process wth rate s exponentally dstrbuted wth average The reverse s also true: f the nterarrval tes of events are exponentally dstrbuted wth average then the event countng process s osson wth rate. Thus, the custoer arrval processes of M/M/* queueng systes are osson. / A osson custoer count and exponentally dstrbuted custoer nterarrval tes are the two sdes of the con. / CS 756 6 3

Saplng osson Arrvals Consder a osson custoer arrval process of wth average rate. Each custoer can be classfed as Type I or Type II, wth probablty p and -p respectvely. Then, the arrval process of Type I custoers s also osson wth average rate p. Lkewse, the arrvals of Type II custoers s osson wth average rate p. CS 756 7 Applcaton We know that the custoer arrvals at a barbershop for osson process wth average rate of custoers per hour. Aong the custoers, 4% are ales and 6% are feales. Then the nterarrval tes of ale custoers are exponentally dstrbuted wth an average rate of 4 per hour. The nterarrval tes of feale custoers are exponentally dstrbuted wth an average rate of 6 per hour. CS 756 8 4

Exercse Consder the router confguraton below. ackets er sec. ort 4% ort Mbps 6% ort Mbps The lengths of arrvng packets are exponentally dstrbuted wth an average of bts. CS 756 9 Questons Argue that queues A and B are ndependent M/M/ systes. Copute the average length of queue A n bts. For a packet destned for port, copute ts expected te at the router ncludng transsson te. CS 756 3 5

Copute the average te a packet spent at the router ncludng transsson te. Copute the average nuber of packets at the routerncludng the ones n transsson. CS 756 3 Mergng osson Arrvals Gven two exponental varables X and X wth rates and, the rando varable X n{ X, X } s also exponental, wth rate +. Consder two osson arrvals, wth average rates and. The erged arrval process wll also be a osson process, wth the average rate + CS 756 3 6

Applcaton Consder the router confguraton below. acket arrval ort ort acket arrval ort Router CS 756 33 The lengths of arrvng packets are exponentally dstrbuted wth an average of bts. Why do we care about packet lengths? acket arrvals at ports and are exponentally dstrbuted wth average rates of and 3 packets per second, respectvely. The transsson rate of port s Mbps. The whole syste can be odeled as a sngle M/M/ queueng syste, wth an arrval rate of 5 and servce rate of,. CS 756 34 7

Burke's Theore In ts steady state, an M/M/ queueng syste wth arrval rate and per-server servce rate produces exponentally dstrbuted nter-departure tes wth average rate. Applcaton: Two cascaded, ndependently operatng M/M/ systes can be analyzed separately. Departs Server Server M/M/ syste M/M/ syste CS 756 35 tfall Consder the syste below where the servers are transsson lnes. 5 ackets er sec. Mbps Mbps ackets lengths are exponentally dstrbuted wth an average of bts. Can the two queues be analyzed separately? Why? CS 756 36 8

Dscusson In general: any feedforward network of ndependently-operatng M/M/ systes can be analyzed n ths syste-by-syste decoposton. p 4 - p 3 CS 756 37 Queston: How about networks that do contan feedbacks? Answer: the nterarrval tes of soe systes ay not be exponentally dstrbuted and thus cannot be analyzed as ndependent M/M/ queues CS 756 38 9

Jackson's Theore For an arbtrary network of k M/M/ queueng systes, where n, n,..., nk n n... k nk j n j n j j j., That s, n ters of the nuber of custoers n each syste, ndvdual systes act as f they are ndependent M/M/ queues they ay not. CS 756 39 Applcaton Consder the network below. The arrval rate and probabltes p and p are known. p CU I / O p We frst copute the arrval rates and : + p, / p, p / p CS 756 4

Let / and /. By Jackson's theore, j, j And N, N Total nuber of custoers n syste s N N + N. Average te n syste s T N +. CS 756 4 Dscusson Consder the packet swtchng network below. Router ackets Router Router 4 Router 3 Can we cte the Jackson's theore, odel the transsson lnes as servers, and analyze the as separate M/M/ queues? Why? CS 756 4