Duality Model of TCP/AQM

Similar documents
Lecture 3: Duality Model of TCP/AQM

Layering as Optimization Decomposition

cs/ee/ids 143 Communication Networks

Network utility maximisation framework with multiclass traffic

Utility Proportional Fair Bandwidth Allocation: An Optimization Oriented Approach

Departure Time and Route Choices with Bottleneck Congestion: User Equilibrium under Risk and Ambiguity

TCP and Network Coding: Equilibrium and Dynamic Properties

Decomposition Methods for Network Utility Maximization

Algorithm Design and Analysis

Internet Congestion Control: Equilibrium and Dynamics

Sliding Mode Based Congestion Control and

Equilibrium of Heterogeneous Congestion Control Protocols

Today. Flow review. Augmenting paths. Ford-Fulkerson Algorithm. Intro to cuts (reason: prove correctness)

IN today networks, error correction is achieved by a

Modeling The Interaction between TCP and Rate Adaptation at Links

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 65, NO. 10, OCTOBER Wenguang Mao, Xudong Wang, Senior Member, IEEE, and Shanshan Wu

Fairness comparison of FAST TCP and TCP Vegas

Control Theory and Congestion

STOCHASTIC GENERALIZED TRANSPORTATION PROBLEM WITH DISCRETE DISTRIBUTION OF DEMAND

The Impact of Imperfect Scheduling on Cross-Layer Rate. Control in Multihop Wireless Networks

Methods for calculation of the coupling coefficients in the Coupling Cavity Model of arbitrary chain of resonators

Chapter Landscape of an Optimization Problem. Local Search. Coping With NP-Hardness. Gradient Descent: Vertex Cover

Understanding TCP Vegas: A Duality Model

M C. Analytical Methods for Network Congestion Control. Steven H. Low Engineering & Applied Science, Caltech

Myriad of applications

Rate Control in Communication Networks

CONGESTION control is a key functionality in modern

Chapter 7. Network Flow. Slides by Kevin Wayne. Copyright 2005 Pearson-Addison Wesley. All rights reserved.

Low order controller design for systems with time delays

On the Stability Region of Congestion Control

Non-linear Control and Observer Design for the Self-energizing Electro-Hydraulic Brake

384Y Project June 5, Stability of Congestion Control Algorithms Using Control Theory with an application to XCP

Improved Approximation of Storage-Rate Tradeoff for Caching via New Outer Bounds

Chapter 10. Closed-Loop Control Systems

Analytical Methods for Network Congestion Control SYNTHESIS LECTURES ON COMMUNICATION NETWORKS

Power Control and Transmission Scheduling for Network Utility Maximization in Wireless Networks

VIII. Addition of Angular Momenta

μ + = σ = D 4 σ = D 3 σ = σ = All units in parts (a) and (b) are in V. (1) x chart: Center = μ = 0.75 UCL =

State Space: Observer Design Lecture 11

Stability analysis of a max-min fair Rate Control Protocol (RCP) in a small buffer regime

Comparison of Augmented State Track Fusion Methods for Non-full-rate Communication

Problem Set 8 Solutions

Research Article Simplicity and Commutative Bases of Derivations in Polynomial and Power Series Rings

Performance Effects of Two-way FAST TCP

PIQI-RCP: Design and Analysis of Rate-Based Explicit Congestion Control

A Theoretical Study of Internet Congestion Control: Equilibrium and Dynamics

Analytic Performance Evaluation of the RED Algorithm

Understanding TCP Vegas: A Duality Model

Analysis and Design of a Proportional-Integral Rate Controller for Streaming Videos

Beyond Cut-Set Bounds - The Approximate Capacity of D2D Networks

Pricing surplus server capacity for mean waiting time sensitive customers

communication networks

Motivation. Lecture 2 Topics from Optimization and Duality. network utility maximization (NUM) problem:

Entropy Minimization in Design of Extractive Distillation System with Internal Heat Exchangers

OSCILLATION AND PERIOD DOUBLING IN TCP/RED SYSTEM: ANALYSIS AND VERIFICATION

CS276 Homework 1: ns-2

Energy Approach-Based Simulation of Structural Materials High-Cycle Fatigue

Gain and Phase Margins Based Delay Dependent Stability Analysis of Two- Area LFC System with Communication Delays

Reliability: Theory & Applications No.3, September 2006

ied widely. The waiting time ditribution and queue ize for a queue with gamma ditributed arrival proce with a pecial Markovian tructure and i.i.d. exp

Control chart for waiting time in system of (M / M / S) :( / FCFS) Queuing model

Effective Capacity-Based Quality of Service Measures for Wireless Networks

The Impact of Imperfect Scheduling on Cross-Layer Rate Control in Wireless Networks

6.3 Buffer Space versus Number of Flows

Analysis of TCP Westwood+ in high speed networks

DTM: A Distributed Traffic Management Algorithm for Sensor Networks

ASSESSING EXPECTED ACCURACY OF PROBE VEHICLE TRAVEL TIME REPORTS

Chapter 4: Applications of Fourier Representations. Chih-Wei Liu

Lecture 21. The Lovasz splitting-off lemma Topics in Combinatorial Optimization April 29th, 2004

Optimal revenue management in two class pre-emptive delay dependent Markovian queues

Packet Size Optimization for Multiple Input Multiple Output Cognitive Radio Sensor Networks aided Internet of Things.

3.1 The Revised Simplex Algorithm. 3 Computational considerations. Thus, we work with the following tableau. Basic observations = CARRY. ... m.

Riser Dynamic Analysis Using WKB-Based Dynamic Stiffness Method

Chapter 7. Network Flow. CS 350: Winter 2018

A New TCP/AQM System Analysis

Relaxation in Glass: Review of Thermodynamics. Lecture 10: Thermodynamic Functions

Congestion Control. Topics

International Journal of Advanced Engineering and Management Research Vol. 2 Issue 5, ISSN:

Chapter 2 Sampling and Quantization. In order to investigate sampling and quantization, the difference between analog

Reliable Data Transport: Sliding Windows

Suggested Answers To Exercises. estimates variability in a sampling distribution of random means. About 68% of means fall

SPREAD: An Adaptive Scheme for Redundant and Fair Storage in Dynamic Heterogeneous Storage Systems

MULTI-PERIOD LOCATION ALLOCATION PROBLEM WITH SAFETY STOCK OPTIMIZATION

NCAAPMT Calculus Challenge Challenge #3 Due: October 26, 2011

On Energy Efficiency of Prioritized IoT Systems: KTH Technical Report

9. Dual decomposition and dual algorithms

<C 2 2. λ 2 l. λ 1 l 1 < C 1

ONLINE APPENDIX FOR HOUSING BOOMS, MANUFACTURING DECLINE,

Optimal Coordination of Samples in Business Surveys

Leopold Franzens University Innsbruck. Responding to Spurious Loss Events in TCP/IP. Master Thesis. Institute of Computer Science

Rate adaptation, Congestion Control and Fairness: A Tutorial. JEAN-YVES LE BOUDEC Ecole Polytechnique Fédérale de Lausanne (EPFL)

ME2142/ME2142E Feedback Control Systems

Demand in Leisure Markets

A Comparative Study of Least-Squares and the Weak- Form Galerkin Finite Element Models for the Nonlinear Analysis of Timoshenko Beams

Structural Control of Probabilistic Boolean Networks and Its Application to Design of Real-Time Pricing Systems

Advanced Digital Signal Processing. Stationary/nonstationary signals. Time-Frequency Analysis... Some nonstationary signals. Time-Frequency Analysis

Throttle Actuator Swapping Modularity Design for Idle Speed Control

FUNDAMENTALS OF POWER SYSTEMS

EP2200 Queueing theory and teletraffic systems

Question 1 Equivalent Circuits

Transcription:

Optimization and Contro of Network Duaity Mode of TCP/AQM Lijun Chen 02/08/2016

Agenda Utiity maimization and dua decompoition An introduction to TCP congetion contro A genera dynamica mode of TCP/AQM The duaity mode of TCP/AQM 2

Network mode A network modeed a a et L of ink with finite capacitie c { c, L} Shared by a et S of ource Each ource ue a ubet L L of ink, which define a routing matri 1 if L R 1 0 otherwie 110 R c 1 c 2 101 2 3 3

Utiity function Each ource attain a utiity U when tranmitting at rate q Non-decreaing function Given the et X of poibe aterna1ve, a func1on U : X R i a u1ity func1on repreen1ng preference rea1on among aterna1ve, if for a y, X, " i at eat a good a y" U U y To ue u1ity func1on to characterize preference i a fundamenta aump1on in economic 4

2 3 Reource aocation probem Each ource attain a utiity U when tranmitting at rate Utiity maimization Key 98 ma. t. U R c 1 1 + 2 c 1 1 + 3 c2 c 1 c 2 110 R 101 1 2 3 c c 1 2 5

Fairne For eatic traffic, U i aumed to be continuouy differentiabe, increaing and tricty concave q Diminihing return U Enure ome kind of fairne Eampe: U 1 α / 1 α, α > 0 q α 1, U og, proportiona fairne y / 0 q α, ma-min fairne 6

ma. t. U R c Poynomia-ovabe, if a the utiity and contraint information i provided. But impractica in rea network Have to eek decompoition to obtain ditributed agorithm 7

8 Lagrangian dua Conider the dua probem Low 99 q Congetion contro: given end-to-end price q Price update: given aggregate ource rate q Price can be updated and fed back to ource impicity c p p R U c R p U p D T c T o p + ma ma : min p R q 1 ' t q U t + + + ] [ 1 c t y t p t p γ R y

Agenda Utiity maimization and dua decompoition An introduction to TCP congetion contro A genera dynamic mode of TCP/AQM The duaity mode of TCP/AQM 9

TCP/IP protoco tack S D appication tranport network ink phyica Doe the part of tak pecific to the particuar appication Provide reiabe end-to-end tranmiion, congetion contro Provide end-to-end path between two end node routing Provide reiabe point-to-point tranmiion, channe acce Provide a ink for tranmitting bit between two node 10

Congetion contro Effect of congetion q Packet o, retranmiion, reduced throughput even congetion coape q Internet ha it firt congetion coape in Oct. 1986 throughput Congetion contro q Achieve high utiization q Avoid congetion q Fair bandwidth haring oad 11

Window-baed fow contro RTT Source 1 2 W 1 2 W time data ACK Detination 1 2 W 1 2 W time q ~ W packet per RTT round trip time q Lot packet detected by miing ACK q Source rate ~ W/RTT 12

TCP congetion contro Source cacuate cwnd from indication of network congetion q cwnd: congetion window ize Congetion indication q Packet oe q Deay q Packet mark Agorithm to cacuate cwnd q Tahoe, Reno, Vega, q DropTai, RED, REM, 13

TCP Reno Jacobon 90 cwnd SS CA time SS: Sow Start CA: Congetion Avoidance 14

Sow tart Start with cwnd 1 ow tart On each uccefu ACK increment cwnd cwnd cnwd + 1 Eponentia growth of cwnd each RTT: cwnd 2 cwnd Enter CA when cwnd > threh 15

Sow Start 1 RTT ender cwnd 1 2 data packet ACK receiver 3 4 5 6 7 8 cwnd cwnd + 1 for each ACK

Congetion Avoidance Start when cwnd threh On each uccefu ACK: cwnd cwnd + 1/cwnd Linear growth of cwnd each RTT: cwnd cwnd + 1 17

Congetion Avoidance ender cwnd 1 data packet receiver 2 ACK 1 RTT 3 4 cwnd cwnd + 1 for each cwnd ACKS

Packet Lo Aumption: o indicate congetion Packet o detected by q Retranmiion TimeOut RTO timer q Dupicate ACK at eat 3 Packet 1 2 3 4 5 6 7 Acknowedgement 1 2 3 3 3 3

Summary: Reno Baic idea q Genty probe network for pare capacity q Draticay reduce rate on congetion for every ACK { if W < threh then W++ ee W + 1/W } for every o { threh W/2 W threh } SS CA 20

TCP Vega Brakmo & Peteron 94 cwnd SS CA time for every RTT { if W/RTT min W/RTT < α then W ++ if W/RTT min W/RTT > α then W -- } for every o W : W/2 queue ize Congetion meaure: end-to-end queueing deay 21

Link agorithm AQM DropTai: drop coming packet when buffer i fu RED random eary detection: warn ource of incipient congetion by probabiiticay marking/ dropping packet q Probabiiticay drop packet q Probabiiticay mark packet marking 1 router B Avg queue 22

Agenda Utiity maimization and dua decompoition An introduction to TCP congetion contro A genera dynamica mode of TCP/AQM The duaity mode of TCP/AQM 23

TCP & AQM p t i t Two component q A ource agorithm: adjut the ending rate baed on congetion Impemented in TCP Tranmiion Contro Protoco q A ink agorithm: update a congetion meaure and end it back to the ource Congetion meaure: o probabiity and deay In form of o/mark or deay Carried out by AQM Active Queuing Management 24

Dynamic mode of TCP/AQM Notation q q q q t : ource rate at time t R : aggregate ource rate at ink t t : ink congetion meaure y p Source can oberve ony t and Link can oberve ony t and t q t R p t : end-to-end congetion meaure of ource q t p t y 25

p t i t q Dynamica mode p t t + 1 + 1 F G p t, t, q y t t q The eact form of F and G are determined by the pecific TCP/AQM protoco 26

Eampe: Vega D d + q for every RTT { if W/RTT min W/RTT < α then W ++ if W/RTT min W/RTT > α then W -- } for every o W : W/2 queue ize F: G: 1 t + t + if w t d t < α d 1 2 D 1 + 1 t 2 D if w t d t > α d t t + 1 t ee p t+1 [p t + y t/c - 1] + 27

Agenda Utiity maimization and dua decompoition An introduction to TCP congetion contro A genera dynamic mode of TCP/AQM The duaity mode of TCP/AQM 28

Duaity mode of TCP/AQM Denote by, p the equiibrium of the ytem p t t + 1 + 1 F G p t, t, q y t t The fied point equation define a reation q f q F > 0 i continuouy differentiabe, and Define a utiity function for each ource U f d, > q Uuay continuou, increaing and tricty concave q Ony determined by tcp agorithm F 0, q F / q 0 impicity 29

30 Eampe: Vega At equiibrium Utiity function d t d t w D t t < + + α if 1 1 2 ee 1 t t + d t d t w D t t > + α if 1 1 2 F: / d q d q d d D d d w α α α α d U og α

Define utiity maimization Dua probem min p o ma. t. R c Interpret ource rate a prima variabe and the congetion price p a dua variabe U D ma U R c p + p T c 31

32 The equiibrium ove the prima and dua, if it atifie KKT condition q The compementary ackne condition i atified by any AQM that tabiize the queue q AQM houd match input rate to capacity to maimize utiization at every botteneck ink, p 0 0 0 ' q U p c y p c y

Remark Revere engineering: the network a an optimization over, and different TCP/AQM protoco a ditributed prima-dua agorithm to ove the utiity maimization and it dua Forward engineering: guide new congetion contro deign q By carefuy chooing utiity function q By propoing better convergent agorithm 33

Can etend to provide a mathematica theory for network architecture and a genera approach to cro-ayer deign appication tranport network ink phyica Minimize repone time web ayout Maimize utiity TCP/AQM Minimize path cot IP Throughput-maima cheduing, Minimize SIR, ma capacitie, 34