PRIME: Peer-to-Peer Receiver-Driven Mesh-Based Streaming

PRIME: Peer-to-Peer Receiver-Driven Mesh-Based Streaming

Nazanin Magharei

Reza RejaieRepresented by: Pavneet Singh Min Luo

Contents of Presentation

Introduction to PRIME Overlay construction in PRIME Content Delivery in PRIME Performance Evaluation of PRIME Conclusions

Introduction to PRIME

Introduction to PRIME

PRIME, a scalable mesh-based P2P streaming mechanism for live content, is based on the concept of P2P.

Introduction to PRIME

A peer-to-peer, commonly abbreviated to P2P, is a distributed network architecture. It composes of participants that make a portion of their resources directly available to other network participants, without the need for central coordination instances.

Introduction to PRIME

This is a diagram of a server-based computer network.

Introduction to PRIME

This is a diagram of a Peer-to-Peer computer network.

Introduction to PRIME

Recently, the success of file swarming mechanisms (e.g., BitTorrent) has motivated another approach to P2P streaming that we call mesh-based P2P streaming.

Swarmstreaming is a technology designed to provide lightning-fast downloads of web content and multimedia without any special server software or special plugins/files. Peers form a mesh-shaped overlay and incorporate swarming (or pull) content delivery.

Introduction to PRIME

Introduction to PRIMEHere is a mesh-based P2P streaming mechanism called SplitStream.

Introduction to PRIME

However, all those mechanisms cannot deliver live content efficiently.

Ⅰ Ensuring the in-time delivery for individual packets of streaming content is difficult.

Ⅱ Since the content is progressively generated by a live source, the availability of new content for delivery is limited.

Introduction to PRIME

So, the PRIME was designed, which is a new mesh-based P2P streaming mechanism for delivery of live content.

The overlay of PRIME

Introduction to PRIME

The delivery pattern contains two phases:

(i) Diffusion phase: Data rapidly flows away from source;

(ii) Swarming phase: Where peers exchange their available packets.

Introduction to PRIME

Introduction to PRIME

As the authors follow a performance-driven approach to design PRIME, in order to measure such performance, two bottlenecks are identified here:

(i) Bandwidth bottleneck: content delivery from its neighbors not fully utilize its incoming access link bandwidth.

(ii) Content bottleneck: there’s not sufficient amount of useful content among its neighbors to effectively utilize its available bandwidth from them.

Overlay construction in PRIME

Overlay construction in PRIME Participating peers in PRIME maintain a

randomly connected and directed overlay. There is a parent-child relationship between

connected peers and content is always delivered from the parent the child.

Each peer maintains connections from multiple parent and serves multiple children. All connections are initiated by children.

Overlay construction in PRIME To construct the overlay, each peer tries to

maintain a sufficient number of parents that can collectively fill its incoming access link bandwidth.

Here, the author roughly estimated the average bandwidth for a connection between parent p to child c as

MIN( )p p c c/outdeg , inbw /indegoutbw

Overlay construction in PRIME MIN( )

Outgoing bandwidth

Outgoing degree of peer p

Incoming bandwidth

Incoming degree of peer c

p p c c/outdeg , inbw /indegoutbwpoutbw

pdegout

cdegincinbw

Overlay construction in PRIME So to avoid a significant bottleneck, we have

the bandwidth-degree condition

i i j j/ deg = / degbwpf outbw out inbw inbwpf infers as bandwidth, or bandwidth-per-flow.This condition implies that all connections in the overlay have roughly the same bandwidth, and this will minimize the bottleneck. This is the first main finding in this paper.

Content delivery in PRIME

Content delivery in PRIME

PRIME incorporates swarming content delivery which combines push content reporting by parents, with pull content requesting by children.

Each peer, as a parent, progressively reports the availability of its new packets to all of its children.

Content delivery in PRIME

In the context of live P2P streaming applications, source progressively generates a new segment of content once every △ seconds. Such segment consists of a group of packets.

So the peer should maintain ω* △ seconds of playout time to swarm among peers.

Content delivery in PRIME

This playout delay between the source and peers has two implications:

(i) Each peer should buffer at least ω* △ seconds worth of content

(ii) Each packet should be delivered within

ω* △ seconds from its generation time to ensure in-time delivery.

Content delivery in PRIME

For the delivery of a single segment, the diffusion phase should first considered.

Peers in level 1 can pull all data units from source during the next interval △ after the new segment is available. It will take

depth * △ seconds until at least one data unit reaches each peer in the system.

Content delivery in PRIME

Diffusion timeDiffusion connectionsDiffusion parentsDiffusion sub-tree

Content delivery in PRIME

Swarming connectionsSwarming parentsPull missing data units fromparents in same or lower levels in different sub-tree

Content delivery in PRIME

For a given overlay, the minimum number of swarming intervals (Kmin) is determined such that a majority of peers can receive their required number of data units of a segment.

As we mentioned that the required buffer contains ω* △ seconds content, we should have

(depth+ Kmin)≤ ω

Content delivery in PRIME

The relevant packets at each scheduling event is divided into 3 sub-windows:

Playing Sub-window Swarming Sub-window Diffusion Sub-window

Content delivery in PRIME

The packet scheduling scheme at each peer is invoked once every △ seconds and takes the following steps:

Quality Adaptation Requesting Diffusion Packets Requesting Playing Packets Requesting Swarming Packets

Content delivery in PRIME

Requesting Swarming Packets

(i) Selecting Timestamp

(ii) Assigning Packets

(1)Description Selection (Random or rarest description

from the useful descriptions among parents)

(2)Parent Selection (Random or based on the minimum ratio

of its assigned packets to its total packet budget)

Content delivery in PRIME

Source can minimize the potential overlap among the delivered content to different diffusion sub-trees and maximize the diffusion rate (the rate of delivery for new packets from source to peers in level 1).

Performance Evaluation of PRIME

Performance Evaluation of PRIME

X axis = Degree, Y axis = % of pop. With quality better than 90%

Degree < 6 : Population is low because of content bottleneck

Degree > 15: Low because of increase in loss rate of individual connections.

Peer Connectivity

Performance Evaluation of PRIME

Performance Evaluation of PRIME X Axis: Percentage of content bottleneck, Y

axis: Percentage of population. Up to degree 12, content bottleneck

increases with a moderate rate with increase in degree due to improved diversity.

Beyond degree 12, increase in content bottleneck due to increase in loss rate is much higher.

Reason: Increase in loss rate.

Performance Evaluation of PRIME This is the second main finding in this paper. There is a sweet range for peer degree over

which swarming content delivery exhibits a good performance and effectively scales with peer population.

The lower bound of this range is 6 but the upper bound is determined by peer bandwidth.

Performance Evaluation of PRIME Loss Rate

X Axis: DegreeY Axis: Bandwidth

Performance Evaluation of PRIME First gap shows bandwidth associated

with lost packets at outgoing access links. Second gap shows bandwidth associated

with lost packets at incoming access links. Aggregate throughput from parent peer to

all of its children's decreases with increase in peer degree.

Performance Evaluation of PRIME

Buffer Requirement

X Axis: DegreeY Axis: Kmin

Performance Evaluation of PRIME Kmin = Required Swarming intervals

As degree increases from 3 to 15, Kmin first decreases because of improved diversity and then becomes constant.

For degree> 15, Kmin increases due to content bottleneck and loss rate of individual connections.

Performance Evaluation of PRIME

The average path length of individual peers decreases with increase in peer degree due to increase in overlay depth.

Pattern of Content Delivery

Performance Evaluation of PRIME

For peer degree 4, difference of 1 hop count. As degree increases, difference in hop count

decreases.

Bi- vs. Uni- Directional Connectivity

X Axis: Average hop count

Y Axis: Perc. Of Pop. With 90% of desirable data.

Performance Evaluation of PRIME This is the third main finding in this paper. The minimum buffer requirement at each

peer is directly proportional to the total duration of the diffusion and swarming phases for each packet.

Bi-directional overlays require larger buffering than Uni-directional overlays.

Performance Evaluation of PRIMEBandwidth Heterogeneity

X axis: Perc. Of content bottleneckY axis: Perc. Of Pop. With 90% of desirable data.

Performance Evaluation of PRIME How is delivery quality affected due to

presence of low bandwidth peers. Minor increase in content bottleneck as

the percentage of parents with high bandwidth decreases.

Content bottleneck also depends upon the position of high bandwidth peers.

Performance Evaluation of PRIME This is the fourth main finding in this paper.

When the amount of resources is sufficient, the heterogeneity and asymmetry of access link bandwidth cannot significantly affects the delivered quality to individual peers.

Performance Evaluation of PRIME

Packet Scheduling

Performance Evaluation of PRIME This graph uses different packet scheduling

algo. The graph suggests that neither criteria for

selecting the description of a packet nor the relative order of selection affects performance.

Scheduling in which packet is requested from a random parent is more likely to experience content bottleneck.

Performance Evaluation of PRIME This is the fifth main finding in this paper.

When load is properly balanced among parents, the performance of delivery has not significantly influenced.

Performance Evaluation of PRIME

As population increases, overlay depth slowly grows, but duration of swarming phase remains constant.

Reason: Increase in population does not increase the no. of diffusion sub trees.

Peer Population

Performance Evaluation of PRIMEPerformance Evaluation of PRIME

Simulation to check the behavior of PRIME when the nodes leave the network.

As the level of distortion increases, the distribution of peer population across different level becomes more imbalanced.

Peer Dynamics

Performance Evaluation of PRIMEPerformance Evaluation of PRIME

Utilization of access link drops with increase in no. of dropped peers.

This proves, both bandwidth and content bottleneck contribute to drop in quality as the overlay distortion increases.

Peer Dynamics

Performance Evaluation of PRIME There are two more main findings in this

paper. One is that incorporating some light weight

coordination mechanism (i.e., careful packet swapping and loss detection) at source can significantly improve overall performance of content delivery.

Performance Evaluation of PRIME The other one is:

The more imbalanced the bandwidth-degree ratio among participating peers, the lower the diffusion rate through the overlay, the lower the delivered quality to individual peers would be.

Conclusion

P2P streaming mechanism can incorporate swarming content delivery.

Bandwidth- degree condition should be satisfied by overlay construction mechanism.