Bandwidth Aggregation in Heterogeneous Networks Kameswari Chebrolu, Ramesh Rao Department of ECE University of California, San Diego.

Slides:



Advertisements
Similar presentations
Multi-Access Services in Heterogeneous Wireless Networks Kameswari Chebrolu, Ramesh R. Rao Abstract Today's wireless world is characterized by heterogeneity.
Advertisements

Streaming Video over the Internet
Scheduling in Web Server Clusters CS 260 LECTURE 3 From: IBM Technical Report.
LYU9903 QoS Schemes in Wireless Networks Ho Pun Mo & Ng Maggie Supervised by Prof. Lyu R.T. Michael 6 December, 1999.
By: Saba Ahsan Supervisor: Prof. Jörg Ott
Computer Networks Performance Metrics Computer Networks Term B10.
Playback-buffer Equalization For Streaming Media Using Stateless Transport Prioritization By Wai-tian Tan, Weidong Cui and John G. Apostolopoulos Presented.
LOGO Video Packet Selection and Scheduling for Multipath Streaming IEEE TRANSACTIONS ON MULTIMEDIA, VOL. 9, NO. 3, APRIL 2007 Dan Jurca, Student Member,
Doc.: IEEE /0604r1 Submission May 2014 Slide 1 Modeling and Evaluating Variable Bit rate Video Steaming for ax Date: Authors:
Fair Real-time Traffic Scheduling over A Wireless Local Area Network Maria Adamou, Sanjeev Khanna, Insup Lee, Insik Shin, and Shiyu Zhou Dept. of Computer.
Quality of Service Issues in Multi-Service Wireless Internet Links George Xylomenos and George C. Polyzos Department of Informatics Athens University of.
Adaptive Video Streaming in Vertical Handoff: A Case Study Ling-Jyh Chen, Guang Yang, Tony Sun, M. Y. Sanadidi, Mario Gerla Computer Science Department,
8.4 WIDE-SCALE INTERNET STREAMING STUDY CMPT 820 – November 2 nd 2010 Presented by: Mathieu Spénard.
Computer Networks Performance Metrics Advanced Computer Networks.
Introduction Future wireless systems will be characterized by their heterogeneity - availability of multiple access systems in the same physical space.
1 On Handling QoS Traffic in Wireless Sensor Networks 吳勇慶.
Generalized Processing Sharing (GPS) Is work conserving Is a fluid model Service Guarantee –GPS discipline can provide an end-to-end bounded- delay service.
Overlay Network and Data Transmission Over Wireless For EE290T Minghua Chen Berkeley.
1 EE 400 Asynchronous Transfer Mode (ATM) Abdullah AL-Harthi.
Leveraging Multiple Network Interfaces for Improved TCP Throughput Sridhar Machiraju SAHARA Retreat, June 10-12, 2002.
End-to-End Analysis of Distributed Video-on-Demand Systems P. Mundur, R. Simon, and A. K. Sood IEEE Transactions on Multimedia, Vol. 6, No. 1, Feb 2004.
Denial of Service Resilience in Ad Hoc Networks Imad Aad, Jean-Pierre Hubaux, and Edward W. Knightly Designed by Yao Zhao.
Real-time smoothing for network adaptive video streaming Kui Gao, Wen Gao, Simin He, Yuan Zhang J. Vis. Commun. Image R. 16 (2005)
In-Band Flow Establishment for End-to-End QoS in RDRN Saravanan Radhakrishnan.
Networking Basics: A Review Carey Williamson iCORE Chair and Professor Department of Computer Science University of Calgary.
Multimedia Applications r Multimedia requirements r Streaming r Phone over IP r Recovering from Jitter and Loss r RTP r Diff-serv, Int-serv, RSVP.
Component-Based Routing for Mobile Ad Hoc Networks Chunyue Liu, Tarek Saadawi & Myung Lee CUNY, City College.
Department of Electronic Engineering City University of Hong Kong EE3900 Computer Networks Transport Protocols Slide 1 Transport Protocols.
CSc 461/561 CSc 461/561 Multimedia Systems Part C: 3. QoS.
On Self Adaptive Routing in Dynamic Environments -- A probabilistic routing scheme Haiyong Xie, Lili Qiu, Yang Richard Yang and Yin Yale, MR and.
1 Networking Basics: A Review Carey Williamson iCORE Professor Department of Computer Science University of Calgary.
Multiple Sender Distributed Video Streaming Nguyen, Zakhor IEEE Transactions on Multimedia April 2004.
Lecture 1, 1Spring 2003, COM1337/3501Computer Communication Networks Rajmohan Rajaraman COM1337/3501 Textbook: Computer Networks: A Systems Approach, L.
Path selection Packet scheduling and multipath Sebastian Siikavirta and Antti aalto.
Tiziana FerrariQuality of Service for Remote Control in the High Energy Physics Experiments CHEP, 07 Feb Quality of Service for Remote Control in.
Packet Scheduling From Ion Stoica. 2 Packet Scheduling  Decide when and what packet to send on output link -Usually implemented at output interface 1.
Mobile IP Performance Issues in Practice. Introduction What is Mobile IP? –Mobile IP is a technology that allows a "mobile node" (MN) to change its point.
Fair Real-time Traffic Scheduling over Wireless Local Area Networks Insik Shin Joint work with M. Adamou, S. Khanna, I. Lee, and S. Zhou Dept. of Computer.
Tiziana Ferrari Quality of Service Support in Packet Networks1 Quality of Service Support in Packet Networks Tiziana Ferrari Italian.
Distributed Multimedia March 19, Distributed Multimedia What is Distributed Multimedia?  Large quantities of distributed data  Typically streamed.
QoS Support in High-Speed, Wormhole Routing Networks Mario Gerla, B. Kannan, Bruce Kwan, Prasasth Palanti,Simon Walton.
Computer Networks Performance Metrics. Performance Metrics Outline Generic Performance Metrics Network performance Measures Components of Hop and End-to-End.
Voice Over Internet Protocol (VoIP). Basic Components of a Telephony Network.
Computer Networks Performance Metrics
Incentive-Oriented Downlink Scheduling for Wireless Networks with Real-Time and Non-Real-Time Flows I-Hong Hou, Jing Zhu, and Rath Vannithamby.
Data and Computer Communications Chapter 11 – Asynchronous Transfer Mode.
Networking Fundamentals. Basics Network – collection of nodes and links that cooperate for communication Nodes – computer systems –Internal (routers,
Analysis of QoS Arjuna Mithra Sreenivasan. Objectives Explain the different queuing techniques. Describe factors affecting network voice quality. Analyse.
Analysis of TCP Latency over Wireless Links Supporting FEC/ARQ-SR for Error Recovery Raja Abdelmoumen, Mohammad Malli, Chadi Barakat PLANETE group, INRIA.
TCP-Cognizant Adaptive Forward Error Correction in Wireless Networks
Challenges to Reliable Data Transport Over Heterogeneous Wireless Networks.
Supporting Multimedia Communication over a Gigabit Ethernet Network VARUN PIUS RODRIGUES.
EE 122: Lecture 15 (Quality of Service) Ion Stoica October 25, 2001.
L Subramanian*, I Stoica*, H Balakrishnan +, R Katz* *UC Berkeley, MIT + USENIX NSDI’04, 2004 Presented by Alok Rakkhit, Ionut Trestian.
Queue Scheduling Disciplines
1 Transport Layer: Basics Outline Intro to transport UDP Congestion control basics.
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.
Downlink Scheduling for Multimedia Multicast/Broadcast over Mobile WiMAX Connection-oriented Multi- state Adaptation Source:IEEE Wireless Communications.
Providing QoS in IP Networks
OverQos: An Overlay based Architecture for Enhancing Internet Qos L Subramanian*, I Stoica*, H Balakrishnan +, R Katz* *UC Berkeley, MIT + USENIX NSDI’04,
Accelerating Peer-to-Peer Networks for Video Streaming
Confluent vs. Splittable Flows
UNIT-V Transport Layer protocols for Ad Hoc Wireless Networks
Computer Science Division
Networking Basics: A Review
Introduction to Packet Scheduling
EECS 122: Introduction to Computer Networks Packet Scheduling and QoS
Modeling and Evaluating Variable Bit rate Video Steaming for ax
Introduction to Packet Scheduling
Presentation transcript:

Bandwidth Aggregation in Heterogeneous Networks Kameswari Chebrolu, Ramesh Rao Department of ECE University of California, San Diego

Introduction Recent mobile Internet growth spurred deployment of different wireless technologies –e.g. GPRS, CDMA2000, HDR, , Bluetooth, Iridium etc End-Users have flexibility regarding Interface choice –Can choose any number of interfaces to best fit application needs Simultaneous use of multiple interfaces opens interesting possibilities –Bandwidth Aggregation, Mobility Support, Security, Reliability Problem Statement: –How to effectively aggregate bandwidth across multiple network interfaces?

Motivation Applications will drive next-generation network deployments Video Applications Video-on-demand Interactive video Video conferencing Multiplayer games –Bandwidth requirements: 250 Kbps to 2-3 Mbps –Problem: Wireless interfaces have bandwidth limitations 50 Kbps – 384 Kbps (GPRS, CDMA2000) TCP applications can also benefit from bandwidth aggregation

Challenges in Bandwidth Aggregation Use of multiple interfaces  Reordering Video applications have stringent QoS requirements –Interactive applications One way latency of 150ms, Max limit 400ms Frame loss rate < 1% –Video on Demand (with VCR functions): One way latency of 1-2 sec Frame loss rate < 1% –Cannot tolerate excess delay due to reordering TCP applications –More than 3 duplicate acks invokes congestion control –Bandwidth probing issues Inter arrival between acks does not reflect available bandwidth

Related Work Link-Layer Solutions –Bonding – aggregates circuit switched lines –IMA – ATM technology for aggregating multiple point-to- point links –Multilink PPP Stripe Protocol –Generic load-sharing protocol based on Surplus Round Robin (SRR) –Minimizes packet processing overhead –SRR similar to WRR Accounts for variable sized packets Surplus (unused bandwidth) is carried on to next round

Related Work (Contd.) Transport-Layer Solutions –RMTP Reliable rate-based transport protocol Flow and congestion control based on bandwidth estimation –Parallel TCP (pTCP) Opens multiple TCP connections on each interface Handles congestion and blackout through data reallocation and redundant striping Network-Layer Solutions –Based on tunneling –Weighted round-robin based scheduling

Research Contibution Solution Approach: –Bandwidth aggregation at IP-level –Meet application requirements using multiple interfaces Contributions: –Architecture for using multiple interfaces based on Mobile-IP –Scheduling algorithm based on estimated delivery time

Outline Architecture Scheduling algorithm Evaluation –Analysis –Trace-based simulation Ongoing work

Outline Architecture Scheduling algorithm Evaluation –Analysis –Trace-based simulation Ongoing work

Architecture for Bandwidth Aggregation Link-Layer Solutions infeasible –End point is an IP address Application/Transport Layer Solutions –Need to modify/rewrite code –Ensure compatibility with existing infrastructure Network Layer solution –IP – a single standard –Application transparency and interoperability –Cleanest Solution

Our Architecture

Architecture Details Mobile IP based –Packets pass through Home Agent (HA) –Simultaneous Binding - multiple Care-of-Address registration –Intelligent scheduling of packets to multiple addresses Radio Access Network Selection Unit (RSU) –Located on Mobile Host (MH) –Selects right interfaces based on app. reqmts. and cost –Update bindings with HA Traffic Management Unit (TMU) –Located on HA and MH –Processes and schedules the incoming traffic onto multiple paths –Conveys application type and end goal requirements to HA Scheduling Algorithm in TMU is crucial –Focus on Interactive Real-Time Applications

Scheduling Algorithm – Design Considerations Bandwidth –Interested in WWAN system (CDMA2000, GPRS etc) Provide only a few hundred kbps –Not interested in WLAN/WPAN systems –Wireless hop is the bottleneck link Delay/Jitter –Wireline Delay – between HA and Base-Station (BS) Delay values and variation small If large, variation may likely be masked at BS as wireless hop is bottleneck –Wireless Delay – between Base-Station and MH Queuing delay and transmission delay

Scheduling Algorithm – Design Considerations Qos Support –Interested in systems that provide QoS (CDMA2000, UMTS etc not HDR) –Negotiated bandwidth and loss rate guaranteed for duration of session

Design Possibility – Weighted Round Robin Schedules packets based on bandwidths of interfaces Not suitable for real-time applications Example: Three interfaces with bandwidth ratios 5:2:1 Packets 1-5 sent on IF1, 6-7 sent on IF2, 8 on IF3 Packet 6 arrives ahead of packets 3,4,5 Packet 3 suffers excess delay due to reordering Ideal ordering: IF1 – 1,2,4,5,6; IF2 – 3,7; IF3 – 8 Variants of WRR – Surplus Round Robin (SRR), Shortest Queue First face similar problems

Our approach: Earliest Delivery First For each path (between HA and MH), estimate arrival time of a packet at MH Estimation based on –Bandwidth of the interface –One-way wireline delay (estimated) on the Internet path Schedule the packet on the path that delivers the packet the earliest Quick remarks –No need for synchronized clocks (relative one-way delay counts) –EDF is not work conserving –EDF cannot totally eliminate reordering –Multiple applications can be handled by combining EDF with Weighted Fair Queuing (WFQ)

EDF Details Each path l is associated with three quantities –A variable, which is the time the channel becomes available next. –, the one-way wireline delay (estimate) of the path –, the bandwidth negotiated - the arrival time, - the size of packet i, Packet i scheduled on path l would be delivered at the MH at EDF schedules the packet on the path p for which is updated to

Performance of EDF How well can EDF perform? –Can the application QoS requirements be met? –Is performance as good as having a Single-Link (SL) with the same aggregated bandwidth? Approach –Analysis Prove fairness of EDF in distributing bits across different links Compare EDF with SL in terms of work, delay, jitter and buffering –Simulation Consider application performance level metrics Measure sensitivity of the algorithm to bandwidth asymmetry, number of interfaces, delay variation, channel losses

Properties of EDF Notation: – - max packet Size, – number of interfaces, - bandwidth of link l, - weight of link l (normalized bandwidth) Assumptions: –, and When packets are of constant size, they arrive in order at the client For variable sized packets – Given P packets to transmit, the maximum difference in normalized bits allocated to any two pair of links is –For WRR, this amount is a function of P and can be unbounded –For SRR it is

Properties of EDF (Contd.) For any time t, the difference between the total number of bits W serviced by SL and EDF is The difference in delay experienced by a packet i in SL and EDF is bounded by The jitter experienced by a packet i without buffering is upper bounded by The jitter experienced by a packet I with buffering is upper bounded by The buffer size needed to deliver the packets in order is

Experimental Methodology Trace driven simulation Server –Video frame traces – office cam (Mpeg4 and H.263) For MPEG-4, avg – 400kbps, peak - 2Mbps, frame period - 40ms For H.263, avg – 260kbps, peak – 1.5Mbps, frame period - variable Maximum packet size assumed is 1400 byte Home Agent –Employs scheduling algorithm Base-Station –No cross traffic –Serve packets first-come-first-serve basis

Experimental Methodology (Contd.) Client –Begin video display after a fixed delay – startup latency L –Afterwards, display frames consecutively every t seconds (frame period) –Arrival after playback deadline results in frame loss –Startup latency bounds one-way delay of packets Internet Path –Packet delay traces collected over different Internet paths –Hosts on UCSD, UCB, Duke, CMU –Wireline delay range used 15ms – 22 ms (one-way) Algorithms under comparison –Single Link – SL –Surplus Round Robin - SRR

Application Performance Metrics Backlog in the system Delay experienced by packets Frame Loss probability - Fraction of packets that miss playback deadline Glitch Duration: Number of consecutive frames that cannot be displayed Glitch Rate: Number of glitches/sec

Bandwidth Allocation % Bandwidth Needed over SL to achieve 0% frame loss, MPEG-4, BS = 3

Backlog SLEDFSRR Backlog in the system between HA and Client application, MPEG-4 Bandwidth fixed at 600kbps

Delay Distribution Cumulative Percentage of Delay, Mpeg-4, BS=3

Frame Loss probability

Sensitivity to Bandwidth Asymmetry

Sensitivity to Number of Interfaces

Extensions to EDF

Other Results Delay Variation : EDF –Truncated Gaussian with mean 22ms, std. devn. 0-10ms –For a split 5:3:1 at 225ms, No variation introduces 0.26% frame loss 5ms variation, 0.27% frame loss 10ms variation, 0.28% frame loss Channel Losses –Limited retransmissions help Other Applications –Non-Interactive Applications Large tolerance for delay  no big difference in relative perf. –Video-On-Demand Applications High peak-to-mean rates imply over-provisioning of bandwidth –Choice of scheduling algorithm does not matter

Summary Network-layer architecture to enable multiple communication paths EDF scheduling algorithm: reduces delay experienced by packets in presence of multi-path. An analysis of the algorithm shows that it doesn’t differ much from idealized SL Trace-driven simulations –EDF mimics SL closely –Outperforms by a large margin WRR based approaches

Ongoing Work Bandwidth Aggregation in Best-Effort Systems –Bandwidth Estimation at MH –Work ahead scheduling TCP –Support TCP applications –Network layer solutions Ad-hoc Networks Security