Connection-Level Scheduling in Wireless Networks Using Only MAC-Layer Information

Similar documents
Markov Chains and Multiaccess Protocols: An. Introduction

ETSN01 Exam Solutions

Optimal Routing and Scheduling in Multihop Wireless Renewable Energy Networks

Interference avoidance in wireless multi-hop networks 1

A Distributed Throughput-Optimal CSMA/CA

June 20th, École Polytechnique, Paris, France. A mean-field model for WLANs. Florent Cadoux. IEEE single-cell WLANs

The Capacity of Wireless Networks

Power Laws in ALOHA Systems

Queuing Delay and Achievable Throughput in Random Access Wireless Ad Hoc Networks

Analyzing Multi-Channel Medium Access Control Schemes With ALOHA Reservation

A Survey on Congestion Control and Maximization of Throughput in Wireless Networks

Introduction: Two motivating examples for the analytical approach

Architectures and Algorithms for Wireless Networks

IEEE/ACM TRANSACTIONS ON NETWORKING 1. On the Efficiency of CSMA-CA Scheduling in Wireless Multihop Networks

Flow-Level Analysis of Load Balancing in HetNets and Dynamic TDD in LTE

CS551 Router Queue Management

Queuing. Congestion Control and Resource Allocation. Resource Allocation Evaluation Criteria. Resource allocation Drop disciplines Queuing disciplines

Throughput-Optimal Scheduling in Multihop Wireless Networks Without Per-Flow Information

Performance Analysis of WLANs Under Sporadic Traffic

TSIN01 Information Networks Lecture 3

Unit 2 Packet Switching Networks - II

Lecture 9 November 12, Wireless Access. Graduate course in Communications Engineering. University of Rome La Sapienza. Rome, Italy

Multiple Access (1) Required reading: Garcia 6.1, 6.2.1, CSE 3213, Fall 2010 Instructor: N. Vlajic

ECEN 5032 Data Networks Medium Access Control Sublayer

Delay-Based Back-Pressure Scheduling in Multihop Wireless Networks

Read Chapter 4 of Kurose-Ross

TCP challenges in multi-hop wireless networks. Why multi-hop?

Lecture 21. Reminders: Homework 6 due today, Programming Project 4 due on Thursday Questions? Current event: BGP router glitch on Nov.

Wireless Networks (CSC-7602) Lecture 8 (15 Oct. 2007)

CHAPTER 5. QoS RPOVISIONING THROUGH EFFECTIVE RESOURCE ALLOCATION

Queue Length Stability in Trees Under Slowly Convergent Traffic Using Sequential Maximal Scheduling

Queuing Systems. 1 Lecturer: Hawraa Sh. Modeling & Simulation- Lecture -4-21/10/2012

Distributed Power Control and Coding-Modulation Adaptation in Wireless Networks using Annealed Gibbs Sampling

Multi-Channel MAC for Ad Hoc Networks: Handling Multi-Channel Hidden Terminals Using A Single Transceiver

CS 556 Advanced Computer Networks Spring Solutions to Midterm Test March 10, YOUR NAME: Abraham MATTA

CHAPTER 3 EFFECTIVE ADMISSION CONTROL MECHANISM IN WIRELESS MESH NETWORKS

WSN NETWORK ARCHITECTURES AND PROTOCOL STACK

Resource Allocation in Contention-Based WiFi Networks

Throughput-Optimal Broadcast in Wireless Networks with Point-to-Multipoint Transmissions

Some Optimization Trade-offs in Wireless Network Coding

Throughput Analysis of CSMA Wireless Networks with Finite Offered-load

Fairness in the IEEE network. Shun Y. Cheung

Modeling Wireless Networks for Rate Control. David C. Ripplinger

B. Bellalta Mobile Communication Networks

TCP Unfairness in Ad Hoc Wireless Networks and a Neighborhood RED Solution

Overview Computer Networking What is QoS? Queuing discipline and scheduling. Traffic Enforcement. Integrated services

Channel-Hopping Multiple Access

cs/ee 143 Communication Networks

Blocking Avoidance in Wireless Networks

Ad hoc and Sensor Networks Chapter 13a: Protocols for dependable data transport

MAC Protocols and Packet Switching

CHAPTER 5 PROPAGATION DELAY

Packet multiple access and the Aloha protocol

Multiple Access. Data Communications and Networking

Fast, Efficient, and Robust Multicast in Wireless Mesh Networks

Stochastic Processing Networks: What, Why and How? Ruth J. Williams University of California, San Diego

Scheduling Algorithms to Minimize Session Delays

Exploring the Inefficiency and Instability of Back-Pressure Algorithms

Analytical Models for Understanding Misbehavior and MAC Friendliness in CSMA Networks

Time-Step Network Simulation

OSI Data Link Layer. Network Fundamentals Chapter 7. Version Cisco Systems, Inc. All rights reserved. Cisco Public 1

CSMA based Medium Access Control for Wireless Sensor Network

ELEC 537: COMMUNICATION NETWORKS, PROJECT REPORT, FALL

Multiple Access Communications. EEE 538, WEEK 11 Dr. Nail Akar Bilkent University Electrical and Electronics Engineering Department

Wireless Multicast: Theory and Approaches

Router s Queue Management

Onroad Vehicular Broadcast

Modeling the protocol under different capture and sensing capabilities

Stream Sessions: Stochastic Analysis

Stop-and-Go Service Using Hierarchical Round Robin

DDSS: Dynamic Dedicated Servers Scheduling for Multi Priority Level Classes in Cloud Computing

Introduction to Real-Time Communications. Real-Time and Embedded Systems (M) Lecture 15

COMP3331/9331 XXXX Computer Networks and Applications Final Examination (SAMPLE SOLUTIONS)

Random Access. 1. Aloha. 2. Slotted Aloha 3. CSMA 4. CSMA/CD

Performance Evaluation of Scheduling Mechanisms for Broadband Networks

Modèles pour le calcul du délai pire-cas de communication dans les réseaux IoT industriels

Markov Model Based Congestion Control for TCP

A Novel Scheduling and Queue Management Scheme for Multi-band Mobile Routers

Throughput of Random Access without Message Passing

PERFORMANCE ANALISIS OF THE LEMR-MULTICHANNEL PROTOCOL

A Better-Than-Best Effort Forwarding Service For UDP

Comparison of pre-backoff and post-backoff procedures for IEEE distributed coordination function

SIMPLE MODEL FOR TRANSMISSION CONTROL PROTOCOL (TCP) Irma Aslanishvili, Tariel Khvedelidze

CapEst: Estimating wireless link capacity in multi-hop networks

Promoting the Use of End-to-End Congestion Control in the Internet

048866: Packet Switch Architectures

On Generalized Processor Sharing with Regulated Traffic for MPLS Traffic Engineering

Basic Switch Organization

Co-existence of WiFi and ZigBee

1. Data Link Layer (Layer 2)

A Multi-channel MAC Protocol for Ad Hoc Wireless Networks

Outline Computer Networking. TCP slow start. TCP modeling. TCP details AIMD. Congestion Avoidance. Lecture 18 TCP Performance Peter Steenkiste

A Distributed CSMA Algorithm for Maximizing Throughput in Wireless Networks

Toward a Reliable Data Transport Architecture for Optical Burst-Switched Networks

Priority Traffic CSCD 433/533. Advanced Networks Spring Lecture 21 Congestion Control and Queuing Strategies

The War Between Mice and Elephants

Wireless Networks (CSC-7602) Lecture 8 (22 Oct. 2007) Seung-Jong Park (Jay) Fair Queueing

GC-HDCN: A Novel Wireless Resource Allocation Algorithm in Hybrid Data Center Networks

Medium Access Control Protocols With Memory Jaeok Park, Member, IEEE, and Mihaela van der Schaar, Fellow, IEEE

Channel-Hopping Multiple Access

Transcription:

Connection-Level Scheduling in Wireless Networks Using Only MAC-Layer Information Javad Ghaderi, Tianxiong Ji and R. Srikant Coordinated Science Laboratory and Department of Electrical and Computer Engineering

Outline Background and Description of Problem Centralized scheduling algorithm with only MAC layer information Distributed implementation using CSMA.

Outline Background and Description of Problem Centralized scheduling algorithm with only MAC layer information Distributed implementation using CSMA.

Wireless Networks: Interference Model A collection of links with interference constraints Users may not be able to transmit simultaneously due to interference. Feasible schedule: a collection of links that can be activated simultaneously Medium Access Protocol (MAC): determines which users are to transmit at each time instant. Must satisfy interference constraints Achieve efficient use of resources Performance metrics: throughput and delay

Wireless Networks: Traffic Model Files/Connections arrive at link l according to some stochastic process with rate λ l. File/Connection size ~ mixture of Geometric distributions There are K file types, each type is geometrically distributed with different mean. A file arrived at links l can belong to type i with probability p li, i=1, 2,.., K. Motivation: Heavy-tail-distribution of file sizes in Internet Most of files are short but most of bytes are generated by long files. By controlling probabilities p li, for a fixed average file size, we can generate distributions with such properties.

MaxWeight: Static Connections Connections are not time varying. Packets of a connection arrive at a link according to some stochastic process. Selects a feasible schedule which maximizes the sum of weights of the selected links Weights are functions of queue lengths

MaxWeight: Dynamic Connections In presence of file/connection arrivals and departures Weight = a function of the total number of packets of all files at link Problem: Links with small queues may starve for long periods of time. Severe unfairness if file sizes can vary widely (short files may starve for long periods of time)

Background and Description of Problem Centralized Scheduling Algorithm with Only MAC-Layer Information Distributed implementation using CSMA.

Layered Network Model Scheduler is implemented as part of MAC layer. The TCP layer controls packet arrivals into the MAC layer Window flow control protocol, window size 1 Questions: Is MaxWeight with only MAC layer information still throughput optimal? Can it resolve the unfairness issue (better delay performance)?

Scheduling algorithm: Network State Which links to activate at each time? Which MAC layer packets to serve at an active link? Decisions should be made only based on MAC layer information. Network State: For each file: MAC-layer queue, congestion window size, transport-layer flow indicator (whether there are still packets of file at Transport layer or not) Additional information required for MAC-layer service discipline (FIFO, random, ), e.g., ordering of MAC-layer packets. Congestion window dynamics determined by the state Network Markov Chain is well-defined.

MaxWeight with MAC layer Information Weight of link l is a function of its MAC queue length (with modification when queue is zero): Our scheduling protocol: At each time t, the Log-Max-Weight algorithm finds a schedule s which maximizes

Throughput Optimality Main Result: The scheduling algorithm maximizes throughput, independent of congestion control algorithm and (non-idling) service discipline. Proof Sketch: The max weight algorithm with weight log(q l ) is throughput optimal, where q l is the expected number of packets waiting at link l. (Note: q l is a function of network state.) The difference between the link weight using q l and the weight using the MAC queue length q l mac is small. Therefore, using the MAC queue length is sufficient to ensure throughput optimality

Throughput Optimality: Proof Sketch n l : number of files at link l σ : max mean file size Because Taking log() The image cannot be displayed. Your computer may not have enough memory

Simulations: File Transfer Delays Only assumption on congestion window sizes used by the transport layer is that they are greater than zero and upper bounded 1-hop Interference model Mixture of Geometric distributions: mean 2 with probability 5/6 and mean 100 with probability 1/6. Simulations for different traffic intensities MAC-layer packets are removed in a FIFO order.

Simulations: File Transfer Delays Delay performance of short files Delay performance of long files Short file : files which are less than half of the mean file size.

Simulations: File Transfer Delays A large total queue length at a link does not imply a large number of files at that link. Regular MaxWeight chooses a link with large queue length but containing only a few files. MAC scheduler yields better delay performance compared to regular MaxWeight. 1-bit additional information to identify short files for shortfile-first MAC service discipline: delay performance will be even better.

Background and Description of Problem Centralized Scheduling Algorithm with Only MAC-Layer Information Distributed Implementation Using CSMA.

Distributed Implementation Previous result holds with weight functions of form where g(.) is an arbitrary increasing function. Use CSMA using such weight functions: Each time slot (t) is divided into two parts Control Data choose a decision schedule m(t) choose a data transmission schedule x(t)

Distributed Implementation Step 1: Generating Decision Schedule m(t) Each link i transmits an INTENT message with probability a i in the control slot t. Those links that transmit INTENT messages and do not hear any INTENT messages from the neighboring links consist a decision schedule m(t). Step 2: Generating Transmission Schedule x(t) For any link i in m(t) do If no links in its conflict set C(i) were active in the previous data slot, link i will decide to become active with probability p i : x i (t)=1 inactive with probability : x i (t)=0 Else, link i will remain inactive: x i (t)=0 Step 3: Data Transmission: In the data slot, use x(t) as the transmission schedule.

Distributed Implementation: Example 1 2 3 4 5 6 7 Current schedule: x(t)={1, 5} Select a decision schedule: m={3, 5, 6} Allowed decisions for links in m: link 3: x 3 =0 (no choice) Link 5: x 5 =0 (w.p. 1-p 5 ) link 6: x 6 =1 (w.p. p 6 ) Other links states unchanged New schedule: x(t+1)={1, 6}

Distributed Implementation x(t) evolves as a Discrete-Time Markov Chain. If the weights do not change with time, the steady-state probability of using schedule x has the following productform: By letting, we have. But queues are actually changing with time, so weights are time-varying. ensures that weights change slowly enough such that the Markov chain remains close to its equilibrium distribution.

Distributed Implementation Hence, distributed algorithm chooses schedule x with probability where. Distributed algorithm chooses the MaxWeight schedule with high probability. Combined with previous result CSMA using only MAC layer information is throughput optimal where (No time scale separation assumption) g(.) can be arbitrarily slow increasing, so for large range of practical queue lengths.

Conclusions Scheduling Using MAC layer information Scheduler is implemented as part of MAC Layer Can resolve the unfairness issue (better delay) Can be throughput optimal if the link s weight is chosen as a log-wise function of its MAC queue. Such a weight function can be used in distributed implementation using CSMA (still throughput optimal with no time scale separation assumption) Under investigation: Extension to Multihop, general file size distributions