Upload
reynold-logan
View
219
Download
1
Embed Size (px)
Citation preview
ON DESIGING END-USER MULTICAST
FOR MULTIPLE VIDEO SOURCES
Y.Nakamura, H.Yamaguchi, A.Hiromori, K.Yasumoto†, T.Higashino and K.Taniguchi
Osaka University†Nara Institute of Science and Technology
ICME2003 2July/9/2003
Goal
Realizing Multi-party video conferencing systems Many-to-many multicast application which consists of
hundreds of users User hosts exchange multiple video streams in real-time
Efficient use of bandwidth is required
Internet
ICME2003 3July/9/2003
Application Layer Multicast (ALM)
ALM is multicast on overlay networks End users act as multicast routers Does not require special hardware such as IP multicast
enable routers Application-specific routing protocols can be designed More efficient than Unicast because a sender does not
need to send data to all receiversUnicast A
B
C
S
D
A
B
C
S
D
ALM
ICME2003 4July/9/2003
Design Issues Each video uses some amount of bandwidth on overlay
networks→In delivering multiple video streams, they compete for bandwidth on overlay links
Users may have priority requirements to video streamse.g. users may prefer the speaker’s video than audience’s
Internet
ICME2003 5July/9/2003
InternetOverlay Network
End-user Multicast for Multi-party Application (Emma)
New ALM protocol for multi-party communication systems Users construct overlay networks Each user host sends its own video continuously and receives some of other
user hosts’ video streams on overlay networks The sending and receiving video streams are controlled based on users’
priority requirement given to the video streams
Red
Red
ICME2003 6July/9/2003
Users’ Join When a new user wants to join a session, the user gets the current
users’ addresses from a central server The user measures delay with these users and establish overlay links
with the closest two or three users to construct mesh-like overlay networks
Existing routing trees are expanded so that they do not overlap each other as long as they can
E C A
BD
Routing tree can be expanded with•small delay from each source•less duplication
AddressManagementServer
A:133.1.xx.xxB:129.3.yy.yyC: …
ICME2003 7July/9/2003
Priority-based Video Delivery Each user specifies a priority requirement as a priority value
to each video stream. A request message with a priority value is forwarded to the
video source user But there is no enough bandwidth on the path
C A
BD
B: 5
A: 8
C: 6
D:15
ICME2003 8July/9/2003
Priority-based Video Delivery In forwarding the message, the followings are calculated step
by step how to stop existing video streams how to minimize the loss of priority values
If the loss of priority value < the gain of priority value by accepting the request The existing streams are stopped and the new stream is delivered
C A
BD
B: 5
A: 8
C: 6
D:15
The sum of the satisfied priority values
becomes larger than before
ICME2003 9July/9/2003
Users’ Leave and Failure Even if an user host leaves, trees can be recovered
When a user leaves, the simplest way to recover tree is to let child nodes connect with the parent of the leaving user
However, this may cause the concentration of overlay links to the parent user To prevent this, each user periodically collects the information of # of overlay links of its
descendant users When a user leaves, its parent can immediately suggest the children of the leaving user
where to connect again
C A
B
G
H F D
EUser E leaves
D,F,GC,D
Recovery can be done quickly without causing
unbalanced trees
ICME2003 10July/9/2003
Performance Evaluation
We have developed a simulator for evaluating performance of EmmaWritten in an object oriented scripting language
Ruby Network model
Hierarchical topology model based on tiers146 nodes, 66 usersUsers give priority values to video by Zipf’s law
ICME2003 11July/9/2003
Users’ Satisfaction
The variation of average of satisfied priority value per userCompared with First-Come-First-Serve (FCFS)
method which accepts user requests in a greedy manner
Simulation Scenario Until 5 units of time, 66 users joined After 21 units of time, 1 user joins or leaves at one time
ICME2003 12July/9/2003
Users’ Satisfaction Emma achieves 1.5 times higher values than FCFS Under the presence of users’ leave and join, priority values
are kept high
0
20
40
60
80
100
120
140
160
180
1 4 7 10 13 16 19 22 25 28 31 34 37 40 43
time
prio
rity
val
ue p
er u
ser
(ave
.)
40
50
60
70
# of
use
rs
# of users FCFS Emma
ICME2003 13July/9/2003
Routing Tree Duplication on Overlay Link
# of routing trees on an overlay linkLess duplication is desirable
Simple estimation of average duplication in the ideal case
# of users × (# of users -1) / # of total overlay links
= about 20
→Measure the # of routing trees on each overlay link
ICME2003 14July/9/2003
Routing Tree Duplication on Overlay Link
80% of overlay links have at most 10 routing trees
0
20
40
60
80
100
120
0 3 6 9 12 15 18 21 24 27 30 33
# of routing trees on overlay link
# of
ove
rlay
link
s
00.10.20.30.40.50.60.70.80.91
cum
ulat
ive
ratio
Emma cumulative
ICME2003 15July/9/2003
Conclusion We have proposed new ALM protocol called Emma
Avoiding resource competition
From the experimental results Higher satisfaction of users than a simple method Even though some users leave from or join to a session,
users’ satisfaction is kept high
We have designed and implemented Java middleware based on Emma
(http://www-tani.ist.osaka-u.ac.jp/software/s2.html)
ICME2003 16July/9/2003