1 Implementing Utility-Optimal CSMA Mung Chiang Princeton University Joint work with Jinsung Lee, Junhee Lee, Yung Yi, Song Chong (KAIST, Korea) Alexandre.

Slides:



Advertisements
Similar presentations
A DISTRIBUTED CSMA ALGORITHM FOR THROUGHPUT AND UTILITY MAXIMIZATION IN WIRELESS NETWORKS.
Advertisements

Submission doc.: IEEE /1409r0 November 2013 Adriana Flores, Rice UniversitySlide 1 Dual Wi-Fi: Dual Channel Wi-Fi for Congested WLANs with Asymmetric.
Winter 2004 UCSC CMPE252B1 CMPE 257: Wireless and Mobile Networking SET 3f: Medium Access Control Protocols.
Maximum Battery Life Routing to Support Ubiquitous Mobile Computing in Wireless Ad Hoc Networks By C. K. Toh.
Wireless MACs (reprise): Overlay MAC Brad Karp UCL Computer Science CS 4C38 / Z25 24 th January, 2006.
David Ripplinger, Aradhana Narula-Tam, Katherine Szeto AIAA 2013 August 21, 2013 Scheduling vs Random Access in Frequency Hopped Airborne.
Predictable Performance Optimization for Wireless Networks Lili Qiu University of Texas at Austin Joint work with Yi Li, Yin Zhang,
XPRESS: A Cross-Layer Backpressure Architecture for Wireless Multi-Hop Networks Rafael Laufer, Theodoros Salonidis, Henrik Lundgren, Pascal Le Guyadec.
Interactions Between the Physical Layer and Upper Layers in Wireless Networks: The devil is in the details Fouad A. Tobagi Stanford University “Broadnets.
Contention Window Optimization for IEEE DCF Access Control D. J. Deng, C. H. Ke, H. H. Chen, and Y. M. Huang IEEE Transaction on Wireless Communication.
Priority Queuing Achieving Flow ‘Fairness’ in Wireless Networks Thomas Shen Prof. K.C. Wang SURE 2005.
1 DOA-ALOHA: Slotted ALOHA for Ad Hoc Networking Using Smart Antennas Harkirat Singh & Suresh Singh Portland State University, OR, USA.
Cross-Layer Optimization for Video Streaming in Single- Hop Wireless Networks Cheng-Hsin Hsu Joint Work with Mohamed Hefeeda MMCN ‘09January 19, 2009 Simon.
MAC Layer (Mis)behaviors Christophe Augier - CSE Summer 2003.
Evaluate IEEE e EDCA Performance Tyler Ngo CMPE 257.
*Sponsored in part by the DARPA IT-MANET Program, NSF OCE Opportunistic Scheduling with Reliability Guarantees in Cognitive Radio Networks Rahul.
IEEE OpComm 2006, Berlin, Germany 18. September 2006 A Study of On-Off Attack Models for Wireless Ad Hoc Networks L. Felipe Perrone Dept. of Computer Science.
Achieving End-to-End Fairness in Wireless Networks Ananth Rao Ion Stoica OASIS Retreat, Jul 2005.
Fair Sharing of MAC under TCP in Wireless Ad Hoc Networks Mario Gerla Computer Science Department University of California, Los Angeles Los Angeles, CA.
Mobility Increases Capacity In Ad-Hoc Wireless Networks Lecture 17 October 28, 2004 EENG 460a / CPSC 436 / ENAS 960 Networked Embedded Systems & Sensor.
Opportunistic Packet Scheduling and Media Access Control for Wireless LANs and Multi-hop Ad Hoc Networks Jianfeng Wang, Hongqiang Zhai and Yuguang Fang.
Isolation of Wireless Ad hoc Medium Access Mechanisms under TCP Ken Tang,Mario Correa,Mario Gerla Computer Science Department,UCLA.
MAC Layer Protocols for Sensor Networks Leonardo Leiria Fernandes.
Selfish MAC Layer Misbehavior in Wireless Networks Pradeep Kyasanur and Nitin H. Vaidya 2005 IEEE Reviewed by Dean Chiang.
DETERMINATION OF THE TOPOLOGY OF HIGH SURVIVAL HF RADIO COMMUNICATION NETWORK Andrea Abrardo.
High Throughput Route Selection in Multi-Rate Ad Hoc Wireless Networks Dr. Baruch Awerbuch, David Holmer, and Herbert Rubens Johns Hopkins University Department.
Wireless scheduling analysis (With ns3) By Pradeep Prathik Saisundatr.
Voice Traffic Performance over Wireless LAN using the Point Coordination Function Wei Supervisor: Prof. Sven-Gustav Häggman Instructor: Researcher Michael.
Distributed resource allocation in wireless data networks: Performance and design Alexandre Proutière Orange-FT / ENS Paris.
Opersating Mode DCF: distributed coordination function
DaVinci: Dynamically Adaptive Virtual Networks for a Customized Internet Jennifer Rexford Princeton University With Jiayue He, Rui Zhang-Shen, Ying Li,
1 Dynamic Adaption of DCF and PCF mode of IEEE WLAN Abhishek Goliya Guided By: Prof. Sridhar Iyer Dr. Leena-Chandran Wadia MTech Dissertation.
A Simple and Effective Cross Layer Networking System for Mobile Ad Hoc Networks Wing Ho Yuen, Heung-no Lee and Timothy Andersen.
Distributed Fair Scheduling in a Wireless LAN
An End-to-end Approach to Increase TCP Throughput Over Ad-hoc Networks Sarah Sharafkandi and Naceur Malouch.
DRAND: Distributed Randomized TDMA Scheduling for Wireless Ad- Hoc Networks Injong Rhee (with Ajit Warrier, Jeongki Min, Lisong Xu) Department of Computer.
1 Core-PC: A Class of Correlative Power Control Algorithms for Single Channel Mobile Ad Hoc Networks Jun Zhang and Brahim Bensaou The Hong Kong University.
Message-Passing for Wireless Scheduling: an Experimental Study Paolo Giaccone (Politecnico di Torino) Devavrat Shah (MIT) ICCCN 2010 – Zurich August 2.
Design and Implementation of a Multi-Channel Multi-Interface Network Chandrakanth Chereddi Pradeep Kyasanur Nitin H. Vaidya University of Illinois at Urbana-Champaign.
MOJO: A Distributed Physical Layer Anomaly Detection System for WLANs Richard D. Gopaul CSCI 388.
Congestion Control in CSMA-Based Networks with Inconsistent Channel State V. Gambiroza and E. Knightly Rice Networks Group
EE 685 presentation Utility-Optimal Random-Access Control By Jang-Won Lee, Mung Chiang and A. Robert Calderbank.
Performance Evaluation of WLAN for Mutual Interaction between Unicast and Multicast Communication Session Author: Aamir Mahmood Supervisor: Prof. Riku.
Architectures and Algorithms for Future Wireless Local Area Networks  1 Chapter Architectures and Algorithms for Future Wireless Local Area.
Michael Schapira Yale and UC Berkeley Joint work with P. Brighten Godfrey, Aviv Zohar and Scott Shenker.
DaVinci: Dynamically Adaptive Virtual Networks for a Customized Internet Jiayue He, Rui Zhang-Shen, Ying Li, Cheng-Yen Lee, Jennifer Rexford, and Mung.
Performance Analysis of IEEE Distributed Coordination Function (DCF) Author : Giuseppe Bianchi Presented by: 李政修 December 23, 2003.
IEEE WLAN.
Access Delay Distribution Estimation in Networks Avideh Zakhor Joint work with: E. Haghani and M. Krishnan.
Planning and Analyzing Wireless LAN
An Efficient Gigabit Ethernet Switch Model for Large-Scale Simulation Dong (Kevin) Jin.
Adaptive IEEE MAC Protocol for high efficiency MC-CDMA WLANs 15. FFV Workshop 21. November 2008 Dr.-Ing. Georgios Orfanos 1G.Orfanos, 15. FFV Workshop,
An Efficient Gigabit Ethernet Switch Model for Large-Scale Simulation Dong (Kevin) Jin.
Improving the scalability of MAC protocols in Wireless Mesh Networks Mthulisi Velempini (Mr.)
DRAND: Distributed Randomized TDMA Scheduling for Wireless Ad-Hoc Networks Injong Rhee (with Ajit Warrier, Jeongki Min, Lisong Xu) Department of Computer.
A Comparison of RaDiO and CoDiO over IEEE WLANs May 25 th Jeonghun Noh Deepesh Jain A Comparison of RaDiO and CoDiO over IEEE WLANs.
Distributed-Queue Access for Wireless Ad Hoc Networks Authors: V. Baiamonte, C. Casetti, C.-F. Chiasserini Dipartimento di Elettronica, Politecnico di.
LA-MAC: A Load Adaptive MAC Protocol for MANETs IEEE Global Telecommunications Conference(GLOBECOM )2009. Presented by Qiang YE Smart Grid Subgroup Meeting.
Optimization-based Cross-Layer Design in Networked Control Systems Jia Bai, Emeka P. Eyisi Yuan Xue and Xenofon D. Koutsoukos.
Z-MAC : a Hybrid MAC for Wireless Sensor Networks Injong Rhee, Ajit Warrier, Mahesh Aia and Jeongki Min ACM SenSys Systems Modeling.
MAC Protocols for Sensor Networks
MAC Protocols for Sensor Networks
Architecture and Algorithms for an IEEE 802
Topics in Distributed Wireless Medium Access Control
High Throughput Route Selection in Multi-Rate Ad Hoc Wireless Networks
If You Can’t Beat Them, Augment Them
The Impact of Multihop Wireless Channel on TCP Performance
Javad Ghaderi, Tianxiong Ji and R. Srikant
Horizon: Balancing TCP over multiple paths in wireless mesh networks
Presentation transcript:

1 Implementing Utility-Optimal CSMA Mung Chiang Princeton University Joint work with Jinsung Lee, Junhee Lee, Yung Yi, Song Chong (KAIST, Korea) Alexandre Proutiere (Microsoft Research UK)

2 Q: Can simple yet optimal (in theory) distributed scheduling be deployed? A: Brainstorming and initial answers

3 Background and Related Work Theory Practice Theory-Practice Gap

4 Scheduling in Wireless Networks  When and How to Activate Links for “Good” Performance?  Interference: Restriction on Simultaneous Link Activation  Single vs. multi hop traffic, saturated vs. unsaturated  Performance: Stability/Delay, Utility/Fairness  Design Freedoms  Centralized vs. Distributed  With Collision vs. Without Collision

5 Literature Taxonomy

6 Related Work: R.A. w. Message Passing  Wang Kar 2005  Lee Calderbank Chiang 2006  Mohsenian Huang Chiang Wong 2008  …

7 Simplicity Driven Design  Message passing undesirable  Not fully distributed  Security breach  Synchronization and coordination  Reduction of effective performance  Questions  Q1. How much can we achieve without passing any message?  Q2. Can we implement it over legacy standard?

8 Related Work: Adaptive CSMA  Jiang–Walrand  Allerton 08  Rajagopalan–Shah  CISS 08  Liu-Yi-Proutiere-Chiang-Poor  Microsoft TR 08  Ni-Srikant  ITA 09  Other related work  Jiang-Liew 08  Marbach-Eryilmaz 08

9 Related Work: Implement Backpressure  DiffQ 2009 Warrier Janakiranman Ha Rhee  A theory motivated heuristic solution  Backpressure-based congestion control  e based MAC prioritization  Implemented in the Linux kernel  Horizon 2008 Radunovic Gkantsidis Gunawardena Key  Backpressure-based multi-path routing  Heuristic solution compatible with and TCP  Implemented between data link and network layer

10 Background and Related Work Theory Practice Theory-Practice Gap

11 CSMA No message passing  Content  Sense (and CA or CD)  Hold the channel  Random back-off 123 Interference graph Example

12 System Model  Saturated single-hop sessions  Symmetric Interference  Objective: Utility Maximization  Alternative: Rate Stability for unsaturated input λ1λ1 λ2λ2 Throughput-region Continuous time model first: Access the channel with Poisson rate Back-off counter: exponential distribution Hold the channel with mean duration Channel holding time: exponential distribution

13 Utility-Optimal CSMA (UO-CSMA) Each link l does the following at slot t: Parameters: (1) V >0 (determines “how optimal”), (2) b(t) (decreasing step size)

14 Performance in Continuous Time Model Theorem. With UO-CSMA and decreasing step sizes,

15  Stochastic Approximation  Having controlled Markov noise Proof: Key Ideas Small for large t Slow time-scale Fast time-scale  averaged, and stationary regime Lemma. O.D.E System with schedules with stationary regime Solving dual of “tweaked” problem V. Borkar, “Stochastic approximation with controlled Markov noise”, Systems and Control Letters, 2006

16 Proof: Key Lemma Lemma. Two systems are asymptotically equivalent, i.e., ODE system Averaging effect Original system (continuous interpolation of )

17 Pictorial Description of Key Lemma ODE system Original system Continuous interpolation t t t equal for large t, i.e., for large t Trajectories with the service rate by CSMA being stationary (i.e., averaged)

18 Extension: General Weight Function Extension for strictly increasing, continuously differentiable W(.)

19 Slotted Time Model  Continuous Model  No collisions  Asymptotically arbitrarily close to optimal  However, in practice  Discrete back-off counters  Collisions are unavoidable  Two Questions  Q1: What is the impact of collisions on efficiency?  Q2: Tradeoff between short-term fairness and efficiency?

20 Analysis Approach  We can build a sequence of systems that converge to the continuous system, e.g., as ² decreases  Gap between discrete and continuous case for a fixed  Use the gap between discrete Bernoulli proc. and Poisson proc.  Why tradeoff between short-term fairness and efficiency? Contention probability Gap btwn. discrete and continuous Channel holding time Efficiency Short-term fairness

21 Efficiency and Short-Term Fairness Average duration during which link l do not transmit successfully Short-term fairness is inversely proportional to channel holding time For a given efficiency gap, Channel holding time grows with the order of  Exponential price of short-term fairness for efficiency Short-term fairness index (also useful for TCP)

22 Background and Related Work Theory Practice Theory-Practice Gap

23 Main Goals  Goal 1  Implement and deploy theory-driven scheduling algorithm UO CSMA on top of conventional hardware  Goal 2  Discover, quantify, and bridge the gap between theory and practice for wireless distributed scheduling in general

24 WiMesh  Open Research Testbed (See Song Chong)  Campus-scale wireless mesh testbed with 55 nodes  Easy programmable platform  PC platform, Linux, FOSS  Common Code [Mesh Router]

25 Common Code  Highway from Theory to Experimentation  Reuse GloMoSim simulator for the real mesh router  Experimentation without modification of simulation code  All protocols over link layer –Overlay MAC layer (over IEEE ) –Even cross-layer protocols  Co-verification: co-simulation and experimentation SimulatorTestbed (Reuse without modification) Theory First step verification Second step verification

26 Framework  KAIST WiMesh Testbed with Common Code  Song Chong and Yung Yi Common Code Applications TCP, UDP, … IP, AODV, DSR, … Overlay user MAC L7 L4 L3 L2.5 GloMoSim Codes SimulationImplementation , CSMA, … Two ray, free space, … Hardware Adaptor WLAN NIC Real physical worldSimulation world L2 L1 L2,L1 Easy and fast verification through simulation and experimentation using Common Code

27 Simulation  Simulation environment  Two ray path loss model  Slotted operation 1 timeslot = 1.6ms  1000byte packet size  Link capacity 5Mbps  SNR based packet reception model  No ACK operation  If collision occurs, it lasts for holding time  Backoff counter can be chosen in [0, CW] randomly

28 Implementation  Features  User space implementation  Overlay MAC  UO-CSMA  Per-link queue structure  Virtual queue update  MAC parameter update  MAC Adaptor  Set e QoS parameters  CW, TxOp and AIFS  MAC  FIFO queue per interface  Transmit actual packets Network Layer UO-CSMA (Overlay) (Substrate) srcQ linkQ To dest a To dest b FIFO Queue Wireless (Radio) MAC Adaptor

29  Indoor deployment: part of WiMesh  10 sender-receiver pairs located in 40mx20m Experimental Space

30 Setup WLAN device Atheros 5212 chipset Utility and W function U(x)=log(x), W(x)=x or loglog(x) PHY a mode 5.745GHz band 6Mbps rate Misc. V = 20,100,500 Constant and diminishing step size FlowSingle-hop session Performance metrics Total throughput (or total utility), Throughput deviation Short-term fairness TrafficSaturated

31 Experiment Results  3 link experiment  flows are used  9 interferes with 2 & 10  Total utility and throughput deviation

32 Key Observations Basic Conclusions:  UO CSMA works almost perfectly in simulation  UO CSMA works well in physical reality and on top of legacy drivers  UO CSMA (in implementation) recovers 80% of the difference between DCF and UO CSMA (in simulation)

33 Holding Time  Changing holding time  Similar throughput  Holding time imperfect  collision avoidance  Degrades short-term fairness Same throughput behvior

34  Queues for 0.01 and diminished by 0.9 every 10s  Both cases have similar performance due to equal queue buildup Stepsizes Flow 9 Flow 10 Decreasing step size Flow 9 Flow 10 Fixed step size

35 Parameter V  Changing V parameter  Throughput  Can achieve up to nearly optimum as V increases  Instantaneous Backlog  Qmin = 0.1, Qmax = 2.3

36 Function W  Changing weight function with fixed step size W(x)=xW(x)=loglog(x) Converges in 80sec Converges in 40sec Small backlog Large backlog

37 Background and Related Work Theory Practice Theory-Practice Gap

38 Where Do the Gaps Come From?  Sensing (sensing range and sensitivity)  Holding (enforcing holding and backing off)  Receiving (SIR based and capture effect)  Asymmetry of interference  Asynchronization of clock  protocol overhead  Common Code architecture overhead TheorySimulationExperimentation Over Gap

39 Theory-Practice Gap 1  Theory-Simulation Gap  Nonetheless, simulation follows theory very well GapTheorySimulation BackoffData-slot basedMini-slot based CollisionNo Yes, and last for holding time

40 Theory-Practice Gap 2  Simulation-Implementation Gap  Overhead from Common Code, but small  Indirect physical information to the overlay MAC  Imperfect queue nullification under the overlay MAC  Packet-by-packet WLAN NIC configuration GapSimulationImplementation Carrier sensing Deterministic, CS range= tx range Probabilistic, CS range <= tx range Interference Symmetric, on-off relation Asymmetric, Probabilistic relation Collision Yes, and last for holding time Yes, with retransmission OverheadNo Common Code overhead

41 Theory-Practice Gap 3  Clean slate – over gap GapClean slateOver HoldingPerfectNot perfect Contention controlBy access prob. By discrete backoff, but only 2^n -1 CW available Transmission typeUser definedUnicast with ACK Synchronization Physically synchronized Asynchronous Overhead Hardware dependent WLAN chipset dependent

42 Samples of Workaround Solutions  Ensuring correct holding time  Reducing high prioritized beaconing  MAC prioritization by AIFS and CWmin,max  NAV option using overhearing of wireless  Still, there is some hole  Making CW=0 feasible  CW only can have a form of 2^n-1  Similar transmission chances with CW=1  Still not perfect holding time execution  Packet-by-packet parameter control  Parameter setting in device driver

43 Next Steps on Implementation  This is an ‘interim report’  Many next steps:  Large-scale network with multi-hop sessions  Scaling up the deployment  Routing needs to be considered  Comparison with congestion controlled  n chipset  Ath9k driver is released  Software upgrade  Ath9k device driver support much more freedom  Latest Linux kernel with mac80211 features  Will make the ‘next report’

44 Implementation-Inspired Theory Qs  Many things theory assumed away  Overhead  Asymmetry  Control granularity  Many things theory modeled simplistically  Imperfect holding and sensing  SIR collision model with capture  Many things theory analyzed loosely  Convergence speed  Transient behavior like queue buildup  Parameter choice

45 Theory-Practice  From Dichotomy to Union: Theory Practice Theory Practice The Princeton EDGE Lab