Maximizing User Gain in Multiflow Multicast Streaming on Overlay Networks

1
Maximizing User Gain in Multi-flow Multicast
Streaming on Overlay Networks

• Y.Nakamura, H.Yamaguchi and T.Higashino
• Graduate School of Information Science and
  Technology,
• Osaka University

2
Research 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
3
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 receivers

Unicast
ALM
A
A
B
S
B
S
C
C
D
D
4
Related works

• Overcast,CAN,RMX
• aiming at an efficient delivery of video stream
  in a large-scale group
• HBM
• aiming at overlay network construction for the
  mobile terminal
• Narada,ALMI,Yoid
• aiming at delivery of single video stream in
  small-scale group
• also target the conference application
• Few research to deliver the two or more streams
  simultaneously and continuously for the video
  conferencing

5
Issues to be considered

• 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
  streams
• e.g. users may prefer the speakers video than
  audiences video

Internet
6
Emma/QoS

• New ALM protocol for multi-party communication
  systems
• Users construct overlay network
• Each user sends its own video continuously and
  receives some of other user hosts video streams
  on overlay networks
• Each user filters and adjusts the quality of
  streams and make the space for a new requested
  stream

Red
Red
7
Overlay network construction

• New user joins the session by constructing the
  overlay link
• Measuring the delay with another user, and
  constructing the overlay link with the number of
  appropriate users
• When the link is constructed, the link capacity
  (number of streams that can be delivered) is
  negotiated and decided
• The participating users enhance the delivery
  routing tree
• The routing tree of the source user is
  constructed with the shortest path tree by
  flooding

Underlying Network
8
Leaving failure management

• Many users frequently leave on the overlay
  network
• Descendant node cannot receive the stream
• We need guarantee the continuance delivery of the
  stream
• Need to make the stream had delivered through the
  left node can immediately delivered from another
  node again
• Each node knows the nodes can deliver the stream
  by periodical message
• If neighbor node leaves, the node is immediately
  reconnected to one of them

9
Loss/Gain-based rate adaptation

• Problem
• Because of the restriction and a decrease of the
  link capacity, all requests of stream cannot be
  accepted
• Solutions
• Simple way
• When the stream cannot be delivered, all requests
  of the stream are not accepted
• Narada
• When it is impossible to deliver the stream in
  the received rate in each user, user reduces the
  rate of the stream so that it can be delivered
• Emma/QoS
• Each user decides that it increases / reduces the
  rate of the stream according to the value of the
  gain obtained by receiving the requested stream
  and cutting the rate of the delivering stream
• Requests are accepted as many as possible

10
User gain function

• Every user defines for each stream
• User gain is added when a unit of bandwidth is
  added to current receiving stream
• We use utility function
• This function shows the priority of the user for
  each unit of bandwidth
• e.g. In the typical streaming, utility tends to
  increase suddenly by the rate of a at least
  necessary quality, and to increase gradually in a
  rate increase after that
• User gain is calculated by the difference of the
  following two values
• Value on utility function of k-1 units
• Value on utility function of added k-th unit

Linear approximation
User gain
Utility function
k-1
k
k denotes certain amount of bandwidth on the
overlay links
11
Protocol operations

• Every node periodically sends messages about user
  gain (negative gain) for each stream
• Negative gain Value of user gain lost in
  descendant nodes when a unit is deprived from
  delivering stream
• Request message is transmitted calculating the
  optimal allocation of each unit
• from user gain and a negative gain that increases
  by the request acceptance when requesting it
• As a result, each user can know and decide which
  stream should be reduced

B
of units of links is 3 Streams
12
Performance evaluations

• We have
• implemented a simulator of Emma/QoS
• simulated in a typical video conferencing
  scenario
• compared with Narada (one of the most popular
  ALMs)
• We examined the following items
• Link stress the number of copies of a single
  packet delivered on a physical link
• Path stretch the ratio of the sum of unicast
  hops of the overlay links between two nodes to
  that on the shortest path on the underlying
  physical network
• User satisfaction ratio the ratio of the sum of
  user gain obtained by each node to that of user
  gains requested by the node
• Variation of user gain

13
Link stress and path stretch

• Compare with the performance of Narada
• Narada assumes the delay between hosts to be
  optimizing metric and constructs mesh-like
  overlay network
• On this overlay network Narada constructs
  shortest path tree
• Emma/QoS has better values than unicast
• Maximum link stress is about 10th of the unicast
• The performance of Narada is not so quite
  different from that of Emma/QoS

Link Stress
Path Stretch
14
Distribution of users satisfaction

• The ratio of the total gain of accepted requests
  to that of all requests
• Requests not accepted at all is much smaller than
  Narada and FCFS (First Come First Serve method)
• The admission control mechanism of Emma/QoS is
  useful to video-conference systems

15
Variation of user gain

• Emma/QoS achieves higher user gain than Narada
• when users join/leave during the session
• when users preferences to stream s are changed
  during the session

16
Conclusion

• We have proposed new ALM protocol called Emma/QoS
• To avoid resource competition, we use
  utility-based admission control in decentralized
  way
• 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