1. Taehyun Kim and Mostafa H. Ammar
A Comparison of Layering and Stream Replication Video Multicast Schemes
Taehyun Kim and Mostafa H. Ammar
Presented By:
Kushagra Saxena
Neel Sinha
Hello Everyone, lets start talking about the paper “A Comparison of Layering and Stream Replication Video Multicast Schemes”
Before we start- Now here are meanings of some terms which are important if we want to proceed with understanding this paper
Routing schemes = The schemes we use to send information from sender to receiver or multiple receivers. -> unicast,broadcast,multicast and selecting the right path.
Multicast = multicast is group communication. where data transmission is addressed to a group of destination computers simultaneously. Multicast can be one-to-many or many-to-many distribution.
Hetrogeneous = not uniform /same in some quality or the other/ bandwidth requirement and all.
Replication/replica = same content

2. Overview Aim of the research paper
Comparison between Replication and Layering
Experiments based on the Comparisons
Results
Conclusion

3. What this paper aims at ?
A structured and systematic comparison of video multicasting schemes.
Only those schemes that deal with the heterogeneous receivers.
Replicated Streams.
Cumulative layering.
Non- cumulative Layering.
There are 2 types of layering = cumulative and non cumulative

4. Aim (Contd.)
‘Layered multicast transmission is superior to the replicated stream multicasting’ – widely believed.
Authors contradict this dogma – bandwidth overhead which is incurred by encoding video stream in layers, cannot be neglected while comparison.

5. Replicated Streams ~ More than one video streams.
Replicated – same contents but with different data rates.
However, receiver subscribes to only one suitable stream.
Examples:
SureStream by RealNetworks.
Intelligent Streaming by Microsoft.
Different receivers might want or run on different bandwidths.
Important point being: Receiver subscribes to om;y 1 suitable stream

6. Replicated Streams (Contd.)

7. Replicated Streams (Contd.)
R1, R2 and R3 are from different domain
Receivers subscribe to only one stream
R1 joins the high quality stream (8.5Mbps)
R2 receives the medium quality stream (1.37Mbps)
R3 joins the low quality stream (128kbps)

8. Cumulative Layering ~
Video can encoded in a base layer and one or more enhancement layers.
Base Layer: Independently decoded.
Enhancement Layer(s): Decoded with lower layers to improve the video quality.
Layer ‘k’ can be only be decoded along with layers 1 to k-1.
Example: MPEG-2 scalability modes.
In this approach, the video is encoded in a base layer and one or more enhancement layers. The base layer can be decoded independently, but the enhancement layers can be decoded cumulatively (i.e., layer can only be decoded along with layers 1 to k-1)
The enhancement layers contribute to the improvement of the video quality that leads to the progressive refinement.

9. Non- Cumulative Layering ~
Video is encoded in two or more independent layers.
Two or more independently decoded layers.
Receivers select any subset of video layer and join it, without joining the layer-1 multicast group.
Eg: Multiple Description Coding.

10. Layered Multicast (Contd.)

11. Layered Multicast (Contd.)
R1 subscribes to all video layers (10 Mbps)
R2 joins enhancement layers 1 and the base layer (1.5 Mbps)
R3 just receives the base layer (128kbps)

12. Layering or Replication ?
Common belief: ‘Layering is better than replication.’ - Really ?
Bandwidth overhead in layering.
Cater to specifics of encoding.
Implicit Protocol Complexity
Topological placement of receivers
We have to look at the ->
a bandwidth overhead is incurred by encoding a video stream in layers.
This overhead can sometimes change the bandwidth efficiency in favor of replicated stream video multicasting.

13. Layering or Replication ? (Contd.)

14. Layering or Replication (Contd.)
Assuming 20% overhead, the data rates contributing to the video quality are 8Mbps, 1.2Mbps and 102.4Kbps
Stream Replication: video quality are 8.5Mbps, 1.37Mbps and 128kbps

15. Overhead in Layered Video ~
Information theoretic results:
Performance of layered coding is not better than that of non-layered coding. Increase the number of layers - significant quality degradation.
Packetization Overhead:
Enhancement layers carry: Picture header, GoP information and Macroblock information.
Protocol Overhead:
Receivers need to manage the multiple subscriptions in layered video.

16. Experimental Evidence ~
Non-layered streams has better video quality
The layering overhead ranges from 0.4% at 27.7dB PSNR to 117% at 23.2dB PSNR
For a good quality video, the overhead is around 20%
Experimental result of the video quality versus data rates for the flower sequence by comparing MPEG-2 SNR scalability and nonscalability mode.
The video quality is measured in peak signal-to-noise ratio (PSNR).
The overhead increases with quality.

17. A Fair Comparison ~
In order to have a meaningful comparison, need to ensure that each scheme is optimal.
Stream Assignment Algorithm: Determine the reception rate of each receiver by aggregating the data rates of the assigned streams
Rate Allocation Algorithm: Determine the data rate of each stream.
Goal: Maximize the bandwidth utilization by each scheme for a given network, a particular set of receivers and given available bandwidth on the network links
In order to compare between Layered and streaming , we use parameters.
Stream assignment - reception rate of each receiver which is calculated by aggregating the data rates of streams
Rate allocation = data rate of each stream
Maximize bandwidth utilization

18. System Model
Model the network by a graph G = (V, E) V is a set of routers and hosts E is a set of edges representing connection links.
n is number of receivers
Isolated rate:
The reception rate of the receiver if there is no constraint from other receivers in the same session
V- router and hosts
E Connection links
Isolated Rate – reception rate no constraint

19. Stream Assignment - Assign receivers as many layers as possible:
Cumulative Layering: Given stream rates αi
- Assign receivers as many layers as possible:
Compute the isolated rates
Assign Σi αi that does not exceed the isolated rate.
We will compare the stream assignment for three schemes
1. Cumaltive layering
2. Stream replication
3. Non cumulative layering
We will Assign stream rates to a receiver till it does not exceed the isolated rate.

20. Stream Assignment (Contd.)
Stream replication
Define δ = {δi | δi ε R+, i =1,…,m}
δi is the data rate of a replicated stream and m is the number of replicated streams
Set of receivers assigned to stream i.
Two objectives
Minimum reception rate for all receivers is greater than zero
Maximum as much as possible.
Greedy algorithm
Allocate δ1 to all receivers to satisfy the minimum reception rate constraint
Receiver is assigned a stream that has not been assigned and has the maximum value of group size and stream rate product
Receiver can either subscribe to base or any other high quality layer.
S = data rate of the replicated stream.
In this scheme , number of receivers are aissnged to a particular stream
We use greedy algorithm in this apporahc
S1 to all receiver
The assignment of streams to the receiver deepends on factors such as whether it has been assigned a, the maximum value of the group size etc
Receiver subsbribe to any layer

21. Stream Assignment (Contd.)
Non-cumulative layering
Define
i is the data rate of a non-cumulatively layered stream and m is the number of streams
Set of receivers assigned to stream i
Two objectives
Minimum reception rate for all receivers is greater than zero
Maximum as much as possible.
Y = data rate of non cumulateve layered streams
We maximize the bandwidth efficiency as much as possible

22. Rate Allocation Cumulative layering A session is defined by
where αi is data rate of layer ‘i’ and ‘m’ is the no. of layers.
The authors used an optimal receiver partitioning algorithm to determine optimal stream rates using dynamic programming
This maximizes overall EFFECTIVE reception rate
We will now compare rate allocation for the three schemes.
Rate allocation represents data rate of the streams.
Ai is the data rate of stream i
Optimal receiver partition algorithm which uses dp is used determine optimal stream rates
It maximizes the reception rate

23. Rate Allocation Stream replication
Stream rates, i, are allocated based on the optimal cumulative layering rate.
1 is the stream rate of the base. If a receiver can join up to k layers, the receiver has the capability to join a replicated stream of data rate k.
Stream rates are allocated based on Optimal cumulative layering rate.
If a receiver can join up to k layers, the receiver can subscribe to a stream of data rate k.

24. Rate Allocation (Contd.)
Non-cumulative layering
Receiver can subscribe to any subset of layers without joining the base layer
 = data rate of non-cumulatively layered stream.
2m-1 different link capacities with m non-cumulative layers
i are allocated based on i =>
Receiver can subscribe to any number of layers . Subscribing to base layer is not necessary.
Y = data rate
yi is allocated on the basis of ai

25. Experiments ( Performance Metrics)~
Average reception rate
Average rate received by a receiver
Average effective reception rate
Amount of data received less the layering overhead
Total bandwidth usage
Adding the total traffic carried by all links in the network for the multicast session
Efficiency
total effective reception rate / total bandwidth usage
In the experiments , we will evaluate the performance matrix in terms of 1.

26. Network Model Georgia Tech Internetwork Topology Models (GT- ITM)
1 server
1640 nodes with 10 transit domains
4 nodes per transit domains, 4 stubs per transit node, 10 nodes in a stub domain
transit-to-transit edges = 2.4Gbps
stub-to-stub edges = 10Mbps and 1.5Mbps
transit-to-stub edges = 155Mbps, 45Mbps and 1.5Mbps
number of layers = 8
amount of penalty = 20%
The networks consists of server, nodes ,transit domains and edges of different kind

27. Experiment Results Random Receiver Distribution - Reception Rate:
Cumulative layering can receive more data
Number of layers in cumulative layering is twice as many as that of non- cumulative layering
Randomly select a server and receivers from a set of nodes in the graph. Receivers are selected from all domains which results in random distribution of receivers
Authors then investigate the performance of the video multicast schemes by varying the number of receivers
Cumulative layering can receive most data

28. Experiment Results
Random Receiver Distribution - Effective Reception Rate:
Stream replication has the highest effective reception rate.
The efficiency of replicated stream is also the best Stream replication has the highest effective reception rate

29. Experiment Results
Random Receiver Distribution - Total Bandwidth usage:
Cumulative layering has the highest Total Bandwidth Usage.
Cumulative layering has the highest Total Bandwidth Usage

30. Experiment Results
Random Receiver Distribution - Bandwidth usage efficiency:
Stream Replication has the highest Bandwidth usage Efficiency.
Stream replication has the highest bandwidth efficiency

31. Experiment Results Clustered receiver distribution.
In Clustered receivers are chosen within only one transit domain and a sender is selected from another domain. Multiple streams share the bottleneck link
layered video multicasting is more efficient than replicated stream video multicasting
the performance of layered video multicasting is improved but that of replicated stream video multicasting is degraded
performance characteristics are changed in favour of layered video multicasting, when receivers are clustered in a small number of domains.

32. Conclusion Conclusion is in line with expectations
Replication is better suited to a RANDOM DISTRIBUTION of hosts
Layering is better suited to a CLUSTERED DISTRIBUTION of hosts
EFFECTIVE reception rate of a layered approach is significantly lower than the channel reception rate.
Replication - Random distribution
Layering – clusetered distribution

33. Questions ??