Transcript Presentation

Real-time Transport for Assured Forwarding:
An Architecture for both Unicast and Multicast
Applications
By Ashraf Matrawy and Ioannis Lambadaris
From Carleton University, Ottawa, Canada
Proc. Of IEEE ICC, 2003
Presented by Fang Yan
12/14/04
agenda
Motivation
Related works
Network model
End-to-end architecture
The rate adaptation algorithm
Simulation results
Conclusion
Motivation
Develop a multicast congestion control
scheme that relies on the IETF proposed
Assured Forwarding(AF) architecture.
AF helps build a simple end-to-end architecture.
AF is expected to be deployed soon in Internet
routers
For simplicity, marking/policing is done at the
senders, instead of the edge routers.
Related works -- AF
A means for a provider to offer different levels of
forwarding assurances for IP packets received
from a customer
Better reliability than best-effort service
Four AF classes are defined, each AF class is
allocated a certain amount of forwarding resources
(buffer space and bandwidth)
Within each AF class IP packets are marked with
one of three possible drop precedence values
Related works -- RED
Random Early Detection
Widely used Active Queue Management
(AQM) technique.
Parameters:
Avg: the average queue size
Minth: the minimum threshold
Maxth: the maximum threshold
RED (contd.)
Algorithm
RED (contd.)
Calculate Pa
Pb = maxp(avg – minth)/(maxth – minth)
pa = pb(1 – count * pb)
Where
Maxp is the maximum value of pb
Count is packets number since last marked pkt
Related work -- RIO
RED with In/Out bits
In/out service allocation profile
congested router preferentially drops out packets
Maintains tow average lengths : in and out
RIO-C : the number of out packets are calculated
based on the total number of packets
RIO-D: the number of out packets are calculated
based on the number of out packets only
Related work -- WRED
Weighted RED
WRED generally drops packets selectively
based on IP precedence
Packets with a higher IP precedence are less
likely to be dropped than packets with a
lower precedence.
Uses one average queue length to make
dropping decisions
Related work -- BECN
Backward Explicit Congestion Notification
uses the existing IP signaling mechanism, the
Internet Control Messaging Protocol (ICMP)
Source Quench (ISQ) message
Congestion notification is kept at the IP level
ISQ are generated by the intermediate congested
RED router and sent back to the source as an
indication of incipient congestion
The source reacts at the transport protocol level by
lowering its data throughput into the network
Network Model
Two-priority queue model
Staggered configuration of class parameters
Routers can send BECN
End-to-End Architecture
Send MPEG4 packets as one multicast group
Packets are marked with different priority level by
the rate adaptation algorithm at the sender
The decisions are based on the congestion status
reported to the sender by the different routers
Congestion status is represented by the probability
of the router sending a BECN message
Always tries to set the rate for the high priority
packets to accommodate the router with the worst
congestion
P(t)
Source R(t)
IP Network
Rate Adaptation Algorithm
Assume that MPEG4 traffic is divided into L
layers marked with L different priorities.
Ri(t), 1<= I<= L, be the rate (in packets/sec)
of layer i at the source at time t.
Pi(t) = PMaxi (t) + PSendi (t) PMinMaxi (t), Pi(t) is
the probability that virtual queue i will
generate a feedback message at time t.
PMaxi (t) = Prob{QueueSize(i) >= max}
PMinMaxi (t) = Prob{min <= QueueSize(i) <= max}
PSendi (t) = Prob{Send feedback message
| min <= QueueSize(i) <= max}
Rate Adaptation Algorithm
Considering the changes from old to
new values of Ri(t) and Pi(t) in a small
interval Δt
Rate Adaptation Algorithm
Δt : the RTT value that corresponds to the
router with the worst situation at the high
priority layer
Routers send a feedback message for packet
that causes a problem with a probability. (2%
~ 5%)
The number of packets between consecutive loss
events is called a loss interval.
Rate Adaptation Algorithm
To calculate Pi(t) at the end of an interval m
K=10
W={4,4,4,4,42,2,2,1,1}
Rate Adaptation Algorithm
Pnewi changes very Δt
At the highest priority layer, take the maximum
At lower priority layers, take the minimum
Subject to Rmini and Rmaxi
Rate Adaptation Algorithm
Simulation setup
Simulation result
high priority throughput
Simulation result
high priority throughput
Simulation result
high priority throughput
Simulation result
low priority throughput
Simulation result
low priority throughput
Simulation result
low priority throughput
Simulation result
total throughput
Simulation result
total throughput
Simulation result
total throughput
Simulation result
packet loss in high priority
Simulation result
packet loss in high priority
Simulation result
packet loss in high priority
Simulation result
packet loss in low priority
Simulation result
packet loss in low priority
Simulation result
packet loss in low priority
Conclusion
 Enables users with different bandwidth capabilities to
receive the same video multicast in different qualities.
 Always try to accommodate the slowest receiver at the
high priority layer
 Allow increasing the rate at the lower priority layer
 RIO-D and WRED result in better utilization of bandwidth,
but also high loss rate in the lower layer
 RIO-C offers different qualities with lower loss rates at the
expense of less bandwidth utilization
Thank you!
Questions?