PowerPoint Template

35 downloads 31736 Views 2MB Size Report
Page ▫ 2. Overview. ▫Introduction. ▫Client-Server Model. ▫Multicast Systems. ▫Peer to Peer. ▫Streaming over the Air. ▫Future / Conclusion ...
Overview and Trends of Video Streaming over IP Aleksandar Markovic and Frank Neugebauer

Overview ƒIntroduction ƒClient-Server Model ƒMulticast Systems ƒPeer to Peer ƒStreaming over the Air ƒFuture / Conclusion

Page ƒ 2

Introduction: Schedule

Page ƒ 3

Introduction: Streaming / IPTV ƒDefinition: Streaming – Transmission of data (audio/video) without the need to complete the transfer prior to playing it.

ƒWhat is IPTV? – Digital Television – IP link, broadband network – Often: Video on Demand (VoD) Page ƒ 4

Media Player ƒDownloading ƒStoring (in the buffer) ƒPlayout – Decompression – Jitter removal – Decoding

Page ƒ 5

Client-Server Model

Page ƒ 6

CDN (1/2)

ƒ Contend Distributed Networks ƒ increased chance for delay and loss ƒ consumption of a large amount of bandwidth Page ƒ 7

CDN (2/2)

ƒ web browser sends a request to the web server ƒ most optimal replicate computed according to client position Page ƒ 8

Multicast: Definition ƒDefinition: – One to many routing topology – Intention to save bandwidth – IPv4

ƒUsage: – Live Video transmission – MBone Page ƒ 9

Multicast: History ƒHistory: – Defined 1980s (RFC 966 and 1112) – 1st transmission 1992

ƒMBone: – Tunneling multicast through unicast networks – Worldwide academic network – Stagnation Page ƒ 10

Multicast: Details ƒStructure: – UDP – Klasse D IP: 1110 + GrouppenID (224.0.0.0 bis 239.255.255.255) – IGMPv1 • Leaving not specified till v2 • v3 as Draft Page ƒ 11

Multicast: Problems ƒDue to UDP: – Unreliable transmission – No congestion control or QoS – Solution: RMTP

ƒBusiness: – Unicast “good enough” – Multicast routers expensive • ISPs see no reason to spend money Page ƒ 12

Multicast: Future (1/2) ƒOutlook: – IPv4: End of lifecycle • Address Space almost consumed – Many unicast architectures (CDN, P2P) are workarounds • Multicast more effective alternative • Internet operates near full capacity: performance increase necessary to avoid “brownout” Page ƒ 13

Multicast: Future (2/2) ƒIPv6: – Multicast already implemented – Consumers become more mobile • Tend to consume more flexible – New Usages: • MMORPG • IPTV • VoD Æ Reduce infrastructure costs Page ƒ 14

Peer-To-Peer Systems ƒ No router support required ƒ Utilization of the application layer ƒ Users act both as clients and servers ƒ Approaches: – tree-based (push) – mesh-based (pull)

ƒ Creation and maintenance of good topology crucial – exchange of data in an efficient, cost-aware manner – delay between peers minimized – number of connections between peers minimized Page ƒ 15

Single-Tree Streaming (1/4) ƒ many possible ways to construct a streaming tree ƒ major considerations: – depth of the tree – fan-out of the nodes

ƒ number of children depends on uploading bandwidth

Page ƒ 16

Single-Tree Streaming (2/4) ƒ Problem: user might leave the streaming session unexpectedly (peer churn) ƒ Descendant peers cut off from receiving data

Page ƒ 17

Single-Tree Streaming (3/4) ƒTree topology: – Centralized » fast recalculation of topology » server possible bottleneck – Decentralized » independence from central server » very slow

Page ƒ 18

Single-Tree Streaming (4/4) ƒ Users grouped into session based on arrival time ƒ Group forms a multicast-tree, denoted as base tree ƒ When a new client joins, then it – must obtain the patch – joins the base tree and retrieves the base stream

Page ƒ 19

Multi-Tree Streaming ƒ Major drawback of single-tree approach: bandwidth of leaf nodes not utilized ƒ Construction of multi-tree system, peers at different positions

Page ƒ 20

Mesh-based systems ƒ freedom of structure overlay ƒ a tracker keeps info about active peers ƒ each peer updates its list about neighbors ƒ basic data unit: video chunk ƒ pulling and buffering of chunks for certain time

Page ƒ 21

Comparison Tree vs. Mesh

Approach

Tree-Based Systems

Mesh-Based Systems

push

pull

Form of exchanged data stream

chunk

Topology

static

dynamic

Maintenance

central (server) or distributed

tracker, exchange of info list

Main problem

peer churn

receiving of chunks not guaranteed

Page ƒ 22

Streaming OTA: Introduction ƒStreaming: – Wired • Slowly reaching maturity – Wireless • New technology • Seamless communication Page ƒ 23

Streaming OTA: Next Gen IPTV ƒIPTV 2.0: – Next Generation Network • Converged network – High bandwidth – QoS – Security – Any User should be able to create or consume content whenever he wants Page ƒ 24

Streaming OTA: Next Gen IPTV ƒIPTV 2.0 (cont’d): – Higher image resolution – On mobile devices – No longer walled garden architecture from ISP – Location based / context sensitive services – Seamless mobility

Page ƒ 25

Streaming OTA: Technology ƒWide Area: – WiMax – W-Lan / Wireless

ƒNear Field: – Bluetooth – ZigBee

Page ƒ 26

Streaming OTA: Wide Area ƒCheap setup ƒMetropolitan areas ƒShared Medium – Good for broadcast – Bad for unicast

ƒProne to errors – FEC – SVC Page ƒ 27

Streaming OTA: Near Field ƒResilient against interference ƒScatter Networks – 50 KB/s – Low resolution video – User generated content – Usefull at gatherings

Page ƒ 28

Streaming OTA: SVC ƒEncoding in layers – Base layer – Enhancement layers

ƒDifferent streams ƒTradeoff – higher video quality – lower efficiency of decoding – Encode once, play everywhere Page ƒ 29

Future: Technology ƒTrends: – Ubiquitous mobile streaming – SVC – IPTV 3.0

Page ƒ 30

Conclusion: Technology ƒNow: – Client-Server – Content Delivery Network – Peer to Peer

ƒSoon: – Multicast

ƒFuture: ƒ Ubiquitous mobile streaming with SVC Page ƒ 31

The End

Any Questions? German and English welcome

Page ƒ 32

Discussion ƒ P2P vs. Multicast: What has more potential? ƒ Where do you see the future of IPTV? ƒ What are advantages/drawbacks of IP Multicast compared to P2P? ƒ In what situations shall the tree-approach be utilized? When is the mesh-approch more suitable? ƒ In tree approach, the coordinating server can become the bottleneck of the system. Is there less probability for the tracker in a mesh-based system to become the bottleneck?

Page ƒ 33