EcnLD, ECN Loss Differentiation to optimize the performance of transport protocols on wireless networks Wassim Ramadan, Eugen Dedu and Julien Bouregois Laboratoire d’Informatique de l’Universit´ e de Franche-Comt´ e
Workshop on Mobile Computing and Networking Technologies 14 October 2009
Wassim Ramadan, Eugen Dedu and Julien Bouregois
EcnLD, ECN Loss Differentiation Method
1 / 20
Introduction Loss differentiation Performance measurements Conclusion
Objective Why DCCP
Outline
1
Introduction Objective Why DCCP
2
Loss differentiation
3
Performance measurements
4
Conclusion
Wassim Ramadan, Eugen Dedu and Julien Bouregois
EcnLD, ECN Loss Differentiation Method
2 / 20
Introduction Loss differentiation Performance measurements Conclusion
Objective Why DCCP
Objective Improving performance of transport protocols over wireless networks Design a new transport protocol suitable for video streaming in wireless networks
Wassim Ramadan, Eugen Dedu and Julien Bouregois
EcnLD, ECN Loss Differentiation Method
3 / 20
Introduction Loss differentiation Performance measurements Conclusion
Objective Why DCCP
DCCP New protocol more adapted for multimedia transmissions : Unreliable Choice between two congestion controls TFRC TCP-like
Possibility to add its own congestion control Mechanisms indicating to the sender with reliability which packets are received by the receiver ECN utilization
Wassim Ramadan, Eugen Dedu and Julien Bouregois
EcnLD, ECN Loss Differentiation Method
4 / 20
Introduction Loss differentiation Performance measurements Conclusion
Objective Why DCCP
DCCP congestion control
TCP-like Similar to the congestion control of TCP But : Packet oriented Selective Acknowledgement (SACK) Well suited to multimedia data transport in environments where there are quick changes in network conditions
Wassim Ramadan, Eugen Dedu and Julien Bouregois
EcnLD, ECN Loss Differentiation Method
5 / 20
Introduction Loss differentiation Performance measurements Conclusion
Motivation ECN Methods based on ECN
Outline
1
Introduction
2
Loss differentiation Motivation ECN Methods based on ECN
3
Performance measurements
4
Conclusion
Wassim Ramadan, Eugen Dedu and Julien Bouregois
EcnLD, ECN Loss Differentiation Method
6 / 20
Introduction Loss differentiation Performance measurements Conclusion
Motivation ECN Methods based on ECN
Cause of losses Congestion Interference, mobility, etc (in Wi-Fi) Why we need to make a distinction between the two causes ? Avoid bad reaction when there is a loss Not to reduce rate to avoid congestion while it is an interference Therefore : maximize throughput transmitted
Classification methods : Three Categories IAT, ROTT or ECN (Our Approach EcnLD).
Wassim Ramadan, Eugen Dedu and Julien Bouregois
EcnLD, ECN Loss Differentiation Method
7 / 20
Introduction Loss differentiation Performance measurements Conclusion
Motivation ECN Methods based on ECN
Cause of losses Congestion Interference, mobility, etc (in Wi-Fi) Why we need to make a distinction between the two causes ? Avoid bad reaction when there is a loss Not to reduce rate to avoid congestion while it is an interference Therefore : maximize throughput transmitted
Classification methods : Three Categories IAT, ROTT or ECN (Our Approach EcnLD).
Wassim Ramadan, Eugen Dedu and Julien Bouregois
EcnLD, ECN Loss Differentiation Method
7 / 20
Introduction Loss differentiation Performance measurements Conclusion
Motivation ECN Methods based on ECN
Cause of losses Congestion Interference, mobility, etc (in Wi-Fi) Why we need to make a distinction between the two causes ? Avoid bad reaction when there is a loss Not to reduce rate to avoid congestion while it is an interference Therefore : maximize throughput transmitted
Classification methods : Three Categories IAT, ROTT or ECN (Our Approach EcnLD).
Wassim Ramadan, Eugen Dedu and Julien Bouregois
EcnLD, ECN Loss Differentiation Method
7 / 20
Introduction Loss differentiation Performance measurements Conclusion
Motivation ECN Methods based on ECN
ECN (Explicit Congestion Notification)
ECN principle Notify the sender without losing packets A packet ECN compatible is marked on a router before its queue to becomes full, otherwise the packet is rejected
Wassim Ramadan, Eugen Dedu and Julien Bouregois
EcnLD, ECN Loss Differentiation Method
8 / 20
Introduction Loss differentiation Performance measurements Conclusion
Motivation ECN Methods based on ECN
EcnLD vs TCP-Eaglet TCP-Eaglet Algorithm : when there are one or more losses, If (Slow Start) : halve transmission rate Else (Congestion Avoidance) And ECN : It is a congestion ⇒ halve transmission rate
Problem : No differentiation in the slow start phase EcnLD, Our approach Use RTT in addition to ECN Algorithm : when there are one or more losses, If ECN OR (n > 0 AND RTTcur > RTTave + RTTvar ) Where : n is the number of losses returned in the acknowledgment It is a congestion ⇒ halve transmission rate Wassim Ramadan, Eugen Dedu and Julien Bouregois
EcnLD, ECN Loss Differentiation Method
9 / 20
Introduction Loss differentiation Performance measurements Conclusion
Motivation ECN Methods based on ECN
EcnLD vs TCP-Eaglet TCP-Eaglet Algorithm : when there are one or more losses, If (Slow Start) : halve transmission rate Else (Congestion Avoidance) And ECN : It is a congestion ⇒ halve transmission rate
Problem : No differentiation in the slow start phase EcnLD, Our approach Use RTT in addition to ECN Algorithm : when there are one or more losses, If ECN OR (n > 0 AND RTTcur > RTTave + RTTvar ) Where : n is the number of losses returned in the acknowledgment It is a congestion ⇒ halve transmission rate Wassim Ramadan, Eugen Dedu and Julien Bouregois
EcnLD, ECN Loss Differentiation Method
9 / 20
Introduction Loss differentiation Performance measurements Conclusion
Simulation topology Simulation results
Outline
1
Introduction
2
Loss differentiation
3
Performance measurements Simulation topology Simulation results
4
Conclusion
Wassim Ramadan, Eugen Dedu and Julien Bouregois
EcnLD, ECN Loss Differentiation Method
10 / 20
Introduction Loss differentiation Performance measurements Conclusion
Simulation topology Simulation results
Network used to perform simulations
The simulation time is 50 seconds. The sender is s1 and the receiver is m1 Wassim Ramadan, Eugen Dedu and Julien Bouregois
EcnLD, ECN Loss Differentiation Method
11 / 20
Introduction Loss differentiation Performance measurements Conclusion
Simulation topology Simulation results
Description of simulation
Objective : compare the performance of EcnLD, TCPlike and TCP-Eaglet Two scenarios with an wireless error rate varying from 0% to 20% Without competition In competition with TCP (between s2 and d1. From 1 to 20s And from 25 to 45s)
One or two MAC retransmissions
Wassim Ramadan, Eugen Dedu and Julien Bouregois
EcnLD, ECN Loss Differentiation Method
12 / 20
Introduction Loss differentiation Performance measurements Conclusion
Simulation topology Simulation results
EcnLD vs TCPlike First scenario : without competition
One retransmission 50000
Two retransmissions 50000
EcnLD, sent packets EcnLD, received packets TCPlike, sent packets TCPlike, received packets
45000 40000
40000
35000
35000 Packets
Packets
EcnLD, sent packets EcnLD, received packets TCPlike, sent packets TCPlike, received packets
45000
30000 25000
30000 25000
20000
20000
15000
15000
10000 5000
10000 0
5
10 Error rate (%)
15
20
0
5
10 Error rate (%)
15
20
Results Improved performance even with increased wireless error rate Wassim Ramadan, Eugen Dedu and Julien Bouregois
EcnLD, ECN Loss Differentiation Method
13 / 20
Introduction Loss differentiation Performance measurements Conclusion
Simulation topology Simulation results
EcnLD vs TCPlike First scenario : without competition
One retransmission 50000
Two retransmissions 50000
EcnLD, sent packets EcnLD, received packets TCPlike, sent packets TCPlike, received packets
45000 40000
40000
35000
35000 Packets
Packets
EcnLD, sent packets EcnLD, received packets TCPlike, sent packets TCPlike, received packets
45000
30000 25000
30000 25000
20000
20000
15000
15000
10000 5000
10000 0
5
10 Error rate (%)
15
20
0
5
10 Error rate (%)
15
20
Results Improved performance even with increased wireless error rate Wassim Ramadan, Eugen Dedu and Julien Bouregois
EcnLD, ECN Loss Differentiation Method
13 / 20
Introduction Loss differentiation Performance measurements Conclusion
Simulation topology Simulation results
EcnLD vs TCPlike Second senario : in competition with TCP
One retransmission 45000
Two retransmissions 45000
EcnLD, sent packets EcnLD, received packets TCPlike, sent packets TCPlike, received packets
40000
EcnLD, sent packets EcnLD, received packets TCPlike, sent packets TCPlike, received packets
40000
35000
35000 Packets
Packets
30000 25000
30000 25000
20000 20000
15000
15000
10000 5000
10000 0
5
10 Error rate (%)
15
20
0
5
10 Error rate (%)
15
20
Results Improving performance even in the presence of other traffic in the network
Wassim Ramadan, Eugen Dedu and Julien Bouregois
EcnLD, ECN Loss Differentiation Method
14 / 20
Introduction Loss differentiation Performance measurements Conclusion
Simulation topology Simulation results
EcnLD vs TCPlike Second senario : in competition with TCP
One retransmission 45000
Two retransmissions 45000
EcnLD, sent packets EcnLD, received packets TCPlike, sent packets TCPlike, received packets
40000
EcnLD, sent packets EcnLD, received packets TCPlike, sent packets TCPlike, received packets
40000
35000
35000 Packets
Packets
30000 25000
30000 25000
20000 20000
15000
15000
10000 5000
10000 0
5
10 Error rate (%)
15
20
0
5
10 Error rate (%)
15
20
Results Improving performance even in the presence of other traffic in the network
Wassim Ramadan, Eugen Dedu and Julien Bouregois
EcnLD, ECN Loss Differentiation Method
14 / 20
Introduction Loss differentiation Performance measurements Conclusion
Simulation topology Simulation results
EcnLD vs TCP-Eaglet First senario : in competition on a wireless network of 11Mb/s
One retransmission 45000
45000
EcnLD, sent packets EcnLD, received packets Eaglet, sent packets Eaglet, received packets
40000
EcnLD, sent packets EcnLD, received packets Eaglet, sent packets Eaglet, received packets
40000
35000
35000
30000
30000
Packets
Packets
Two retransmissions
25000
25000
20000
20000
15000
15000
10000
10000 0
5
10 Error rate (%)
15
20
0
5
10 Error rate (%)
15
20
Results Performances are nearly equal
Wassim Ramadan, Eugen Dedu and Julien Bouregois
EcnLD, ECN Loss Differentiation Method
15 / 20
Introduction Loss differentiation Performance measurements Conclusion
Simulation topology Simulation results
EcnLD vs TCP-Eaglet First senario : in competition on a wireless network of 11Mb/s
One retransmission 45000
45000
EcnLD, sent packets EcnLD, received packets Eaglet, sent packets Eaglet, received packets
40000
EcnLD, sent packets EcnLD, received packets Eaglet, sent packets Eaglet, received packets
40000
35000
35000
30000
30000
Packets
Packets
Two retransmissions
25000
25000
20000
20000
15000
15000
10000
10000 0
5
10 Error rate (%)
15
20
0
5
10 Error rate (%)
15
20
Results Performances are nearly equal
Wassim Ramadan, Eugen Dedu and Julien Bouregois
EcnLD, ECN Loss Differentiation Method
15 / 20
Introduction Loss differentiation Performance measurements Conclusion
Simulation topology Simulation results
EcnLD vs TCP-Eaglet Second senario : in competition on a wireless network de 54Mb/s
One retransmission 80000
80000
EcnLD, sent packets EcnLD, received packets Eaglet, sent packets Eaglet, received packets
70000
EcnLD, sent packets EcnLD, received packets Eaglet, sent packets Eaglet, received packets
70000
60000
60000
50000
50000
Packets
Packets
Two retransmissions
40000
40000
30000
30000
20000
20000
10000
10000 0
5
10 Error rate (%)
15
20
0
5
10 Error rate (%)
15
20
Results EcnLD has a high ratio of received/sent packets TCP-Eaglet has a higher throughput but losses a lot of packets on the network Wassim Ramadan, Eugen Dedu and Julien Bouregois
EcnLD, ECN Loss Differentiation Method
16 / 20
Introduction Loss differentiation Performance measurements Conclusion
Simulation topology Simulation results
EcnLD vs TCP-Eaglet Second senario : in competition on a wireless network de 54Mb/s
One retransmission 80000
80000
EcnLD, sent packets EcnLD, received packets Eaglet, sent packets Eaglet, received packets
70000
EcnLD, sent packets EcnLD, received packets Eaglet, sent packets Eaglet, received packets
70000
60000
60000
50000
50000
Packets
Packets
Two retransmissions
40000
40000
30000
30000
20000
20000
10000
10000 0
5
10 Error rate (%)
15
20
0
5
10 Error rate (%)
15
20
Results EcnLD has a high ratio of received/sent packets TCP-Eaglet has a higher throughput but losses a lot of packets on the network Wassim Ramadan, Eugen Dedu and Julien Bouregois
EcnLD, ECN Loss Differentiation Method
16 / 20
Introduction Loss differentiation Performance measurements Conclusion
Simulation topology Simulation results
EcnLD vs TCP-Eaglet : Loss classification percentage One retr. 11Mb/s EcnLD Eaglet Two retr. EcnLD Eaglet
0% 100% 100% 0% 100% 100%
4% 79% 84% 4% 76% 50%
8% 68% 86% 8% 63% 50%
12% 68% 90% 12% 51% 79%
16% 67% 89% 16% 72% 51%
20% 69% 88% 20% 81% 54%
Avg 72% 86% Avg 73% 64%
One retr. 54Mb/s EcnLD Eaglet Two retr. EcnLD Eaglet
0% 40% 4% 0% 41% 3%
4% 53% 4% 4% 34% 2%
8% 64% 10% 8% 25% 2%
12% 75% 26% 12% 66% 9%
16% 73% 33% 16% 48% 10%
20% 72% 41% 20% 53% 16%
Avg 64% 18% Avg 43% 6%
Results EcnLD has a higher loss classification rate in most cases TCP-Eaglet bad classification results in a higer throughput but less network friendly Wassim Ramadan, Eugen Dedu and Julien Bouregois
EcnLD, ECN Loss Differentiation Method
17 / 20
Introduction Loss differentiation Performance measurements Conclusion
Simulation topology Simulation results
EcnLD vs TCP-Eaglet : Loss classification percentage One retr. 11Mb/s EcnLD Eaglet Two retr. EcnLD Eaglet
0% 100% 100% 0% 100% 100%
4% 79% 84% 4% 76% 50%
8% 68% 86% 8% 63% 50%
12% 68% 90% 12% 51% 79%
16% 67% 89% 16% 72% 51%
20% 69% 88% 20% 81% 54%
Avg 72% 86% Avg 73% 64%
One retr. 54Mb/s EcnLD Eaglet Two retr. EcnLD Eaglet
0% 40% 4% 0% 41% 3%
4% 53% 4% 4% 34% 2%
8% 64% 10% 8% 25% 2%
12% 75% 26% 12% 66% 9%
16% 73% 33% 16% 48% 10%
20% 72% 41% 20% 53% 16%
Avg 64% 18% Avg 43% 6%
Results EcnLD has a higher loss classification rate in most cases TCP-Eaglet bad classification results in a higer throughput but less network friendly Wassim Ramadan, Eugen Dedu and Julien Bouregois
EcnLD, ECN Loss Differentiation Method
17 / 20
Introduction Loss differentiation Performance measurements Conclusion
Conclusion and perspectives
Outline
1
Introduction
2
Loss differentiation
3
Performance measurements
4
Conclusion Conclusion and perspectives
Wassim Ramadan, Eugen Dedu and Julien Bouregois
EcnLD, ECN Loss Differentiation Method
18 / 20
Introduction Loss differentiation Performance measurements Conclusion
Conclusion and perspectives
Conclusion and perspectives
Conclusion EcnLD has a very high rate of received packets, which designed to improve performance on wireless networks EcnLD carries a very high packets reception rate, which makes it suitable to streaming multimedia
Perspectives Improving our contribution in wireless networks to design a new multi-radio protocol
Wassim Ramadan, Eugen Dedu and Julien Bouregois
EcnLD, ECN Loss Differentiation Method
19 / 20
Introduction Loss differentiation Performance measurements Conclusion
Conclusion and perspectives
Conclusion and perspectives
Conclusion EcnLD has a very high rate of received packets, which designed to improve performance on wireless networks EcnLD carries a very high packets reception rate, which makes it suitable to streaming multimedia
Perspectives Improving our contribution in wireless networks to design a new multi-radio protocol
Wassim Ramadan, Eugen Dedu and Julien Bouregois
EcnLD, ECN Loss Differentiation Method
19 / 20
Introduction Loss differentiation Performance measurements Conclusion
Thank you for your attention Questions ?
Wassim Ramadan, Eugen Dedu and Julien Bouregois
EcnLD, ECN Loss Differentiation Method
20 / 20