Transcript ppt

PATH DIVERSITY WITH FORWARD
ERROR CORRECTION SYSTEM FOR
PACKET SWITCHED NETWORKS
Thinh Nguyen and Avideh Zakhor
IEEE INFOCOM 2003
AGENDA





Motivation
System Description
Redundant Path Selection
Simulation Results
Conclusion
MOTIVATION
(1)
(2)
(3)
Packet loss and end-to-end delay
Solutions –
Layered, error-resilient video codecs
Retransmission
Use of edge architectures
Assume single fixed path between sender and
receiver.
PERFORMANCE OF FEC


Depends heavily on the correlation of packet
loss between multiple paths.
So the question is – Are there sufficient
disjoint paths between a pair of senders and
receivers???
MOTIVATION

Theoretical Results –
RS(30,23), packet size-500 bytes, sending rate-800 kbps
Optimal rate partition using two paths.
MOTIVATION
Ratio of irrecoverable loss probabilities of the uni-path scheme to
multi-path scheme.
SYSTEM ARCHITECTURE
SYSTEM ARCHITECTURE Contd.




Sender first executes traceroute from itself to
all the participating nodes and receiver.
It also sends setup packet to all participating
nodes instructing them to execute traceroute.
Sender Computes optimal redundant path
After redundant path is choosen, sender
sends the setup packet to the selected relay
node instructing it to forward packets to the
receiver.
SYSTEM ARCHITECTURE Contd.

Setup Packet = flow ID + IP Address + Port Number

Use UDP to send all the packets between
nodes.
REDUNDANT PATH SELECTION
To measure latency, bandwidth and packet
loss –
(1) Use of Probing tools – inc complexity
(2) Passive Probing
Assumption made in this system –
Two paths with absolute lowest loss rates is
not needed.

REDUNDANT PATH SELECTION



-
-
Sender runs this algorithm.
Get information using traceroute tool. Input
is names of the routers and roundtrip
delays.
Network Topology –
G(V,E) .
Weights denote the latencies b/w
participating nodes.
REDUNDANT PATH SELECTION
Relay node is computed as(1) Compute a set of nodes that result in
minimum number of joint links between the
default Internet path and all redundant
paths.
(2)
Choose k’ that results in minimum weight
associated with the corresponding
redundant path.
SIMULATION RESULTS
Simulation Topologies
Used Brite s/w to generate Alber-Barabasi topologies.

Flat Topology
Two level Hierarchial Topology
SIMULATION RESULTS
To estimate average latency, hop counts and
degree of disjointness –
(1) Choose a set of participating nodes
randomly.
(2) Choose a pair of sender and receiver.
(3) Use the redundant path selection algorithm
proposed to find redundant and default
paths for the given configuration.
Default path – low latency.
SIMULATION RESULTS
(1) Percentage of shared links between redundant and default
paths
SIMULATION RESULTS
(2) Latency of redundant path over the latency of default path
SIMULATION RESULTS
(3) Number of hops of the redundant path over the number hops of the default
path
SIMULATION RESULTS
(4) Cumulative distribution of shared links for various network topologies
SIMULATION RESULTS

NS SIMULATIONS:
Link capacity – 2 Mbs
Packet Loss Rate is considered in 3 scenarios –
(1)
Sender streams the video to the receiver at 800kbps on the
default path.
(2)
Sender streams the video to the receiver on both paths at
400kbps each.
(3)
Same as (2) with a shared link between both paths.
SIMULATION RESULTS
Sending Packets using traditional default path
SIMULATION RESULTS
Both redundant and default path without shared link
SIMULATION RESULTS
Both redundant and default path with one shared link
SIMULATION RESULTS
Avg. loss rate of using one path over both with various number
of shared links between them.
CONCLUSION
-
Simulations show that only 10% of participating
nodes are required for the proposed path
redundant selection scheme to effectively find a
redundant path.
-
DRAWBACKS OF THE PROPOSED SYSTEMPerformance depends on the information given by
traceroute.
Some ASes do not report information about their
networks.
(1)
(2)
REFERENCES
Path Diversity for Enhanced Media Streaming.
John G. Apostolopoulos and Mitchell D Trott, Streaming Media Systems Group.
Thank You