Previously, on CS5248..

1
Previously, on CS5248..
2
IP Multicast
3
New Model IP Multicast
sender
receivers
4
Question 1
should I forward this packet to my subnet?
Router
?
G
5
IGMP v2.0

• JOIN messageA I want to join group G.
• QUERY messageR Which group have you joined ?

6
IGMP v2.0

• LEAVE messageI want to leave group G
• Group-Specific QueryAnybody else belongs to
  group G ?

7
Rate Adaptations
8
Layered Video
Layer 1
Layer 2
Layer 3
9
TCP-Equation
Window size behavior in TCP/IP with constant loss
probability T. Ott, J. Kemperman, and M.
Mathis June 1997, HPCS 1997
10
Heterogeneity
11
Internet Heterogeneity
2 Mbps
56kbps
40kbps
12
Two Simple Solutions

• one stream to fit them all
• send multiple streams

13
(No Transcript)
14
Receiver-Driven Layered Multicast

• Steve McCanne, Van Jacobson, Martin Vetterli
• SIGCOMM 1996

15
Layered Multicast

• 1 Layer 1 Multicast Group
• Receiver subscribes to as many layers as desired

16
RLM Example
S
R1
R2
R3
R4
17
Questions

• How many layers are enough?

18
Solution Join Experiment

• highest layer 1
• join layer 1
• while no packet loss
• highest layer
• join next layer
• leave highest layer
• highest layer --

19
Details

• Tjoin
• Time between join experiments
• Tdetect
• Time taken to detect packet loss

20
Effects of Tjoin

• Need to converge to the right level quickly
• Tjoin should be small
• Repeated failed experiments congest networks
• Tjoin should be large

21
Adapting Tjoin(i)

• One Tjoin per layer
• if join experiment for layer k fails Tjoin(k)
  Tjoin(k)2

22
Example
4
3
2
1
23
Adapting Tdetect

• Set Tdetect to large initial value
• Estimate Tdetect with mean and deviation
• Measure time between join and packet loss occur

24
Two Problems

• Interference
• Scalability

25
Problem 1 Interference
26
Problem 1 Interference
I see, layer 2 is bad for me..
27
Problem 2 Scalability

• Lots of receivers
• Lots of experiments
• Lots of congestions

28
Solution Shared Learning
I am joining layer 2, do not disturb!
29
Solution Shared Learning
30
Solution Shared Learning
I am joining layer 3, do not disturb!
31
Solution Shared Learning
I see, layer 3 is bad for me..
32
Solution Shared Learning
I see, layer 3 is bad for me..
33
Problem

• Reducing interference decrease convergence time
• Solution

34
Evaluation
Convergence Time (s)
100
60
40
20
Session Size
2
100
35
Evaluation
Loss Rate ()
1
Session Size
100
36
Problems with RLM

• Not TCP Friendly
• Share learning is clumsy
• Failed join experiments is bad for the network

37
Thin Streams An Architecture for Multicasting
Layered Video

• L. Wu, R. Sharma., B. Smith
• NOSSDAV 97

38
How bad is failed experiment?

• R sending rate of a layer
• Tj IGMP join latency
• Tl IGMP leave latency
• Tdetect time before drop
• Buffer space at the router

39
How big is IGMP Leave Latency?
40
What To Do?

• Buffer space at the router
• R(Tj Tl Tdetect)
• Proposal of ThinStream
• Reduce R
• Reduce Tdetect

41
Reducing Tdetect

• Detect congestion before it happens

42
Calculating A E

• R bandwidth of one layer
• I measurement interval
• N number of bytes received in I
• G number of layers joined
• A
• E

43
ThinStreams Algorithm

• if (E - A gt leave_threshold) leaveelse if (at
  least Tjoin since last join)
• and (E - A lt join_threshold)
• join

44
Fairness

• Improve fairness
• leave_threshold as a function of G

45
Problems with ThinStream

• Small R
• Still not TCP friendly

46
TCP-Like Congestion Control for Layered Multicast
Data Transfer

• L. Vicisano, J. Crowcroft, L. Rizzo
• INFOCOM 98

47
The RLC Protocol

• Try to emulate TCP
• Synchronize join
• Probe before join

48
Recall TCPs Behavior
AIMD
49
Recall Layered Multicast
4
3
2
1
50
1. Emulating TCP

• Idea By finding appropriate Bi and Tjoin(i)

51
Idea
4
3
2
1
52
2. Synchronizing Join

• Idea send special packets in data stream to
  signal join
• Distance between synchronization point for layer
  i is Tjoin(i)

53
Sync Point
4
3
2
1
54
3. Probe before Join

• Emulate joining the next layer
• If this causes packet drop, dont join

55
Sender-Based Probing
layer i
base layer
56
Sender-Based Probing
layer i
base layer
57
Problems with RLC

• Burst too short
• No fine grain increment

58
Summary

• Receiver Driven Layered Multicast
• Lots of Problems
• No Perfect Solutions