P2P Internet Video Broadcast

8/3/2019 P2P Internet Video Broadcast

http://slidepdf.com/reader/full/p2p-internet-video-broadcast 1/52

Wednesday, April 20, 2011






Unicast




Unicast




Multicast




Broadcast




“Video Broadcast is a simultaneous videodelivery to a large number of receivers”




“The AOL broadcast of Live 8 concert in July 2005, which at the peak has 175,000 simultaneous viewers,and the CBS broadcast of the NCAA tournament inMarch 2006, which at the peak has 268,000

simultaneous viewers. Even with today’s low bandwidthInternet video of 400 Kbps, the CBS/NCAA broadcastneeded more than 100Gbps server and networkbandwidth. As a comparison, Akamai, the largest

commercial CDN service provider, reports a peakaggregate capacity of 200Gbps with its tens of thousands of servers.”



http://slidepdf.com/reader/full/p2p-internet-video-broadcast 9/52Wednesday, April 20, 2011



File Download

• Real time performance

• Bandwidth/latency

• Segment order• Degradation of quality

Audio/Video Conferencing

• Latency

• Multi point

• Small Scale

On demand Streaming

• Asynchronous users




Category Bandwidth-Sensitive Delay-Sensitive Scale

File Download ✘ ✘ Large

On-DemandStreaming

✔ ✔ Large

Audio/VideoConferencing ✘ ✔ Small

VideoBroadcast

✔ ✔ Large






Router based architecture: IP Multicast

Non-Router based architectures

•Peer-to-Peer Architecture




Problems:

• Exhibits performance - complexity tradeoff

•Maintains per-group state

• High level features are more difficult toimplement

•

High cost of bandwidth required for serverbased solutions or CDNs

• Calls for changes at the infrastructure leveland installing multicast capable routers




Advantages:

• Moving the functionality to end systems

• All packets are sent as unicast

• Maintains the “stateless” nature of the

network

• High level features can be significantlysimplified




Disadvantages:

• Redundant traffic

•Increasing latency




Infrastructure Centric Architecture(Content DeliveryNetworks CDNs)

• An organization deploys proxies at strategic locations on

the Internet.

• End systems attach themselves to nearby proxies, andreceive data using plain unicast.

Application end point architecture (P2P broadcast)

• Administration, maintenance, responsibility for theoperation of such a peer-to-peer system are distributedamong the users, instead of being handled by a singleentity.




Infrastructure centric:

• Smaller number of groups, support to tens of thousandsof high bandwidth applications is unclear.

• Potentially more robust data delivery through dedicated,more reliable proxies.

Application end point:

• Instantaneous deployment, minimal setup overhead andcost.

• Potentially enable ubiquitous deployment. However,autonomous users may fail or leave at will.






Characteristics:

• Large scale

• Performance demanding

•Real time constraints

• Degradable quality




Issues that must be addressed:

• Overlay efficiency

• Scalability and load balancing

• Self organizing

• Honor per-node bandwidth constraints

• System Considerations




Tree-based Approach

Data driven Approach




Characteristics:

• Nodes make a tree-like structure rooted at the video source.

• Each node receives video data from its parent and sends

them to its children.

• This approach is push-based.

• Does not require sophisticated video coding algorithms.

Problems

• Poor performance due to failure of nodes.

• Leaves’ out-going bandwidth is not being utilized.




Characteristics:

• Does not construct and maintain an explicit structure for delivering data.

• Uses gossip algorithms

Advantages:

• Resilience to random failures

• The potential bandwidth can be fully utilized.

• The overlay is robust to failures

Problems:

• Redundancy with the high-bandwidth video.

• High startup and transmission delay.




Group Management:

• ESM node maintains information about a small random subset of members, as well as information about the path from the source to itself.

• A new node joins the broadcast by contacting the source and retrieving arandom list of members and then selects one of these members as its parent using the parent selection algorithm.

• To learn about members, a gossip-like protocol is used. Each node A periodically picks one member (say B) at random, and sends B a subsetof group members that A knows, along with the last timestamp it has

heard for each member.

• When B receives a membership message, it updates its list of knownmembers.

• Finally, a member is deleted if its state has not been refreshed in a period.




Membership Dynamics:

•In case a member leaves, members continueforwarding data for a short period, while itschildren look for new parents using the parent selection method.

• Members also send periodic control packets to their children to indicate live-ness.




Performance Aware Adaptation:

• If the node’s performance is significantly below thesource rate, then it selects a new parent.

• The detection time indicates how long a node must stay with a poor performing parent before it switches toanother parent.(for ESM default = 5 sec)

• Switching to a new parent requires going through a slow-

start phase, which may take 1 - 2 seconds to get the fullsource rate.

• The protocol may need to adaptively tune the detection time .




Parent Selection:

• When a node A joins the broadcast, or needs to make a parent change, it probes a random subset of nodes it knows.

• Each node B that responds provides information about:

(i) Performance it is currently receiving, and delay from the source;

(ii)Whether it is degree-saturated or not

(iii)Whether it is a descendant of A. The probe also enables A to determine the round-trip time to B.

• A waits for responses for a timeout period of 1 second.

• From the responses A receives, it eliminates its descendants and members that are saturated.

• For each node B that has not been eliminated, A evaluates the performance it expects to receive if B were chosen as a parent.

• History of past performance is maintained so if A has previously chosen B as parent, then it has anestimate of the bandwidth of the path between B and A.

• If the bandwidth to nodes is not known, then A picks a parent based on delay.




Characteristics:

• The source encodes the stream into sub-streams anddistributes each sub-stream along a particular

overlay tree.

• The quality experienced by a receiver depends on the number of sub-streams that it receives.

Advantages:

• The overall resiliency of the system is improved

• The potential bandwidth of all nodes can be utilized







A CoolStreaming node consists of three keymodules:

(1)Membership manager

(2)Partnership manager

(3)Scheduler




Group and Partner Management:

• Like ESM, CoolStreaming requires newly joining nodes tocontact the origin server to obtain an initial set of partnercandidates.

• CoolStreaming employs an existing Scalable GossipMembership protocol, to distribute membership messages.

• A video stream is divided into segments of a uniform length,and the availability of the active segments in the buffer of a

node is represented by a Buffer Map (BM).

• Each node continuously exchange its BM with its partners,and then determines which segment is to be fetched from which partner accordingly.




Scheduling Algorithm:

• Timely and continuous segment delivery is crucial to video broadcast, but not to file download.

• In CoolStreaming, the playback progresses of the peers are semi-synchronized,and any segment downloaded after its playback time will be useless.

• CoolStreaming adopts a sliding window of 120 segments, each of 1-second video. A BM thus consists of a bitstring of 120 bits, each indicating theavailability of the corresponding segment.

• The sequence number of the first segment in the sliding window is recorded byanother two bytes, which can be rolled back for extra long video programs.

(a)buffer of BitTorrent(b)CoolStreaming




Scheduling Algorithm:

• Given the BMs of a node and its partners, a schedule is then generatedfor fetching the expected segments from the partners.

• For a homogeneous and static network, a simple round-robinscheduler may work well, but for a dynamic and heterogeneousnetwork, a more intelligent scheduler is necessary.

• The scheduling algorithm strikes to meet two constraints:

1. The playback deadline for each segment.

2. The heterogeneous streaming bandwidth from the partners.

(a)buffer of BitTorrent(b)CoolStreaming




Failure Recovery and Partnership refinement:

• The departure of a node can be easily detected after anidle time and an affected node can quickly react throughre-scheduling using the BM information of the remaining partners.

• CoolStreaming also let each node periodically establishnew partnerships with nodes randomly selected from itslocal membership list. This operation serves two purposes:

1. It helps each node maintain a stable number of partners in the presence of node departures.

2. It helps each node explore partners of better quality







Hybrid Approach

Incentives and Fairness

Access Bandwidth Scarce Regimes

Extreme Peer Dynamics and Flash Crowd

Support for Heterogeneous Receivers

Network Coding

Implementation Issues




Chunckyspread hybrid Approach

• Splits a stream into distinct slices and transmitsover separate but not necessarily disjoint trees.

• The participating nodes also form aneighboring graph, and the degree in the graphis proportional to its desired transmission load.

• This hybrid design greatly simplifies the treeconstruction, and largely retains its efficiencyand achieves fine-grained control overload.




In some peer-to-peer broadcast systems, a small set of nodes are requested to contribute 10 to 35 times moreuploading bandwidth than downloading bandwidth.

Designing incentive mechanisms for video broadcastapplications is more challenging than traditional filedownload applications, due to the real-time requirements.

A micro-payment mechanism may be a good solution that

enables video broadcast users to cooperate.

The design of a scalable, light-weight incentivemechanism which can be incorporated into videobroadcast application remains an open problem.




The contribution or upload bandwidth from nodes must exceed the bandwidth that nodes can receive.

The Resource Index (RI) is the ratio of the number of receivers that themembers in the group could potentially sustain to the number of receivers in thegroup for a particular source rate.

• An RI < 1: not all participating nodes can receive the full source rate

• An RI = 1: the system is saturated

• As the RI gets higher, the environment becomes less constrained and itbecomes more feasible to construct a good overlay tree.

frameworks for application-level adaptation:

• Multi-tree framework, where not all nodes receive the full bandwidth.

• The amount of bandwidth a receiver is actually entailed to depends on the total contribution that it makes.




Flash crowd

• There is a large increase in the number of users joining the broadcast in a short periodof time.

High churn

• Users arrive and depart at a very high rate, in which case the peer-to-peer broadcast systemhas to continue to adapt with the peerdynamics.




Layered Video:

• Generates a stream consisting of multiple layers.

• A receiver, depending on its capability, can subscribe to the base layer only with the basic playbackquality, or subscribe to additional layers that progressively refine the reconstruction quality.

MDC (Multiple Descriptive coding):

• Generates multiple streams (descriptions) for the source video.

• Any subset of the descriptions, can be used to reconstruct the video.

• A simple implementation of MD coding can be achieved by splitting even and odd numberedframes.

• The descriptions are then distributed over multiple paths, preferably disjoint, to enhance robustnessand to accommodate user heterogeneity.

Problems:

• Low efficiency because of the iterative motion estimation and transform for all the layers.

• Bandwidth penalty due to transporting the layers.

• Reduce the compression efficiency.

• High computation power to assemble and decode multiple layers.




The fundamental insight in network coding is that if data can be encoded in intermediate

nodes then the optimal throughput can beachieved.

Network coding enhances the robustness,

adaptability, and data availability of a peer-to- peer overlay, because the information isevenly spread in the coded data blocks.




NATs and firewalls

Transport Protocol

Startup delay and Buffer interaction:




NATs and firewalls impose restrictions onconnectivity of nodes on an overlay, and may

prohibit direct communication with one another.

Whether communication is possible between twonodes depends on several factors

• The transport protocol (UDP or TCP)

• The particular kind of NAT/firewall

• Whether the nodes are located behind the same private network.




?= for certain kinds of NATs/firewalls

* = in the same private network




Real implementations have employed TCP as the transport protocol being readily available, widely tested, and may often be engineered to

work well.

The complicating factor in the choice of the transport protocol is they may have different

levels of penetration of NATs and firewalls,and may be treated differently by variousenterprise policies.




A new peer may spend 10 to 15 seconds to join a peer-to-peer overlay, and take another 10 to 15seconds to launch the media player and buffercertain data.

The delay can be significantly longer for someunpopular channels, and will be further prolonged if using TCP and network coding.

The existing peer-to-peer broadcast systemsgenerally separates the streaming engine and the playback engine. An efficient use of this 2-stagebuffer might deliver better performance.







Video broadcast

Difference between video broadcast and other p2p applications

Architectural choices of internet broadcast

• Router based vs. Non Router based

• P2P Architecture

• Infrastructure Centric vs. End point

P2P video broadcast

• Tree-based Approach (ESM, Multi-Tree)

• Data driven Approach (CoolStreaming)

Technical challenges and open issues




“Challenges and Solutions in Peer-to-peer

Live Video Streaming” Masoud Moshref,Hamid R. Rabiee, and Saeed Nari

“Opportunities and Challenges of Peer-to-

peer Internet Video Broadcast”JiangchuanLiu, Sanjay G. Rao, Bo Li, and Hui Zhang




Thank you.Questions?

P2P Internet Video Broadcast

Documents