Reza Rejaie CIS Dept. @ UO1 Prof. Reza Rejaie Computer & Information Science University of Oregon reza Fall 2002 Multimedia

Reza RejaieCIS Dept. @ UO 1

Prof. Reza RejaieComputer & Information Science

University of Oregonhttp://www.cs.uoregon.edu/~reza

Fall 2002

Multimedia Networking Seminar


Class Information

Pre-requisites: CIS532 Class structure• Read assigned papers and submit a review• A designated student presents assigned papers• Participate in class discussions

Goals of the seminar:• To become familiar with research issues,

proposed interesting ideas, design and evaluation methodologies in Multimedia Networking

• To evaluate previous work and identify interesting open research problems in this area


Class Information (cont’d)

Grading P/NP Office hours: W 9-10 Send me an email an specify your presentation topic by Oct 7th. Class schedule!


Format of Presentation/Reviews

What is the problem? Why is it important? What are the assumptions? What is the proposed solution? Which part of the solution is new/interesting? What is the evaluation methodology? What are the interesting & new findings? What are the main contributions and key “take away” points? Leave unimportant details out.


Prof. Reza RejaieComputer & Information Science

University of Oregonhttp://www.cs.uoregon.edu/~reza

Fall 2002

An Overview of Research Problems on Internet Multimedia Networking


Motivation

Rapid growth of multimedia streaming • Popularity of the Web and the Internet• High-bandwidth access (Cable, DSL, LAN)

Poor and inconsistent quality of streams• Small picture size• Low frame rate• Fluctuation in quality

High overhead imposed on the Internet• Long, high-bandwidth streams• Unfriendly to traditional TCP traffic


Internet: An Overview

Heterogeneous network in any way• Protocols, routing, links, network technologies,

end-hosts, bandwidth, delay, etc

Best effort service• Available BW is unknown and variable• Loss rate and loss pattern are unknown and

variable

Resources are shared• Data sources should be well-behaved, i.e. adjust

their transmission with network load Internet applications should be congestion

controlled for network stability and inter-protocol fairness.


Streaming Applications

Encoding mechanisms (e.g. MPEG) are used to compress audio/video signalsAverage BWenc depends on:• Encoding algorithm• Desired quality

Tradeoff between compression efficiency and loss resiliencyAverage BWenc is rather constant but BWenc could be burstyRealtime constraint

Encoder

Decoder

A/V Source

Display

IPB

BWenc

quality


Supporting Streaming Applications

Mismatch between streaming applications’ requirements and best-effort service Alternative solutions:• Add new services to the network

• Integrated Services (IntServ)• Differentiated Services (DiffServ)

• Enable applications to cope with best-effort service

• Adaptive streaming applications

Encoder

Decoder

Display

BWenc

quality

BWave


Adaptive Streaming Applications

CC should be integrated Important losses should be repaired/recoveredQuality adaptation matches stream BW with avg. available BW• To avoid buffer

over/underflow

Can we use TCP?

Cong. Ctrl

Buffer

Decoder

Server

Display

Encoder

Source

TCPTCPInternet

QualityAdapt


Role of Quality Adaptation

BW(t)

Time

Mismatch between ave. BW and stream BW• Buffering absorbs

short-term mismatch• Long-term mismatch

triggers quality adaptation

Frequent variations in quality is disturbing

Ave. BW

Stream BW


Delay sensitivity of Streaming App.

The higher the level of interactions, the higher the sensitivity to delay, the lower the amount of buffered data

• Internet telephony, Video conferencing• Video games• Live but non-interactive (i.e. lecture-mode)• On-demand playback of stored video

Less

Inte

ract

ive


Issues with Client-Server Streaming

Congestion control (CC) Loss repair (LR) Quality adaptation (QA)


CC for Streaming

Goal: to achieve a fair share of network BW among co-existing flows• TCP-friendly CC is desired

CC should probe for excess BW and reduce tx rate when congestion is detectedConflicting requirements:• Smooth adjustments in BW are desired• Responsive to changes in network load

Sender-driven vs Receiver-driven CC Rate-based vs Window-based


Loss Repair (Error Control)

Primary approaches:• Retransmission (Retx)• Forward Error Correction (FEC)

Retx has higher BW efficiency but results in longer recovery delay => More buf.• Appropriate for non-interactive app

FEC has shorter recovery delay but lower BW efficiency => Less buf.• Media-dependent vs media-independent

Effect of pkt loss on perceived quality is encoding-specific


Quality Adaptation

Adaptive encodingSwitching between multiple encodingTrans-coding or Selective Frame DroppingLayered encodingMultiple Description encoding

Effect of quality adaptation on perceived quality is encoding-specific


Protocols & Other Design Issues

Protocols:• RTP/RTCP• RTSP

Packetization• Application Level Framing (ALF)


Limitations of Client-Server Arch.

Limited scalabilityLimited & unstable qualityAsynchronous access could be inefficientRTT could be high• High delay VCR-functions• Large startup delay

Increasing network capacity doesn’t solve these problems?

Content should be close to interested clients

Server

ClientClientClient

Internet

Server

server

Client Client Client

Internet


New Distribution Architectures

Extending client-server architecture• Proxy Caching• Content Distribution Networks (CDN)

Replacing client-server architecture• Peer-to-Peer Networks


Proxy Caching for Streaming Media

Client

Client

Client

Client

Client

Server1Internet

Server2

Client

Client

Proxy

Proxy

ISP

Campus


CDN for Streaming Media

Client

Client

Client

Client

Client

Server1Internet

Server2

Client

Client

Server1

Server1

ISP

Campus


Proxy Caching & CDN: Design Issues

Idea: Bring popular contents close to clientsGoals: Minimizing network load, Maximizing delivered quality Main design issue: content-management

1) Push or Pull a popular streaming object?2) Which stream to push/pull & When?

• How to determine popularity/value of an object?

3) Which replacement algorithm?• Granularity of replacement, e.g. replacing part of an

unpopular stream or the entire stream?

4) Which quality of a stream to cache/load?


Peer-to-peer Networks for Streaming

Client

Client

Client

Client

Client

Server1Internet

Client

Client

ISP

Server2


P2P streaming: Design Issues

Idea: No central server, each active peer who has an object could act as a server for that object. Goals: localize traffic, and balance load among active peers Main design issues:• How to locate an object?• Which object to cache/replace at a peer?

No server => new way of thinking• e.g. streaming a MD. encoded stream from

multiple peers


Measurement & Characterization

Measurement is a useful method to gain more insight about certain behavior/phenomenon• e.g. Behavior of aggregate traffic at a server,

user access pattern to a server• Measurement often leads to characterization

of the behavior/phenomenon under investigation

• Characterization of a behavior allows us to improve design/performance of a related network component (e.g. protocols/server)

Main issues: Data collection & Data analysis (and both are challenging)


Measurement for Streaming Media

Sample measurement issues:1) What is the characterization of user

access pattern to mm objects at a server, a proxy, or a peer (in a p2p network)?

2) What is the characterization of mm objects on the Web (e.g. length, bw, etc)?

3) What kind of CC, QA or LR mechanisms (if any) existing streaming applications implement?


Summary of Topics

Congestion control (2 sessions) Quality adaptation (1/2 session) Loss repair (1 session) Multimedia proxy caching (1 session) CDN for streaming (1/2 session) MM traffic measurement (2 sessions)

You should send me an email to select a topic for your presentation by Oct 7th .

Documents

Reza Rejaie CIS Dept. @ UO1 Prof. Reza Rejaie Computer & Information Science University of Oregon reza Fall 2002 Multimedia