1. Improving TCP over Wireless by Selectively Protecting Packet Transmissions Carla F. Chiasserini Michele Garetto Michela Meo Dipartimento di Elettronica Politecnico di Torino, Italy

2. Presentation outline Introduction to the problem Solutions proposed in the literature Our approach Simulation results Conclusions …

3. Introduction: TCP over Wireless Fundamental problem TCP assumes all losses due to congestion Wireless breaks this assumption: losses due to channel errors, handoffs … and TCP unnecessarily reduces window, resulting in low throughput and high latency  Improving TCP over wireless is a must !

4. Proposed schemes to improve the performance of TCP over wireless Adaptive Forward Error Correction (AFEC) Snoop protocol TCP-SACK Airmail Explicit Loss Notification (ELN) SMART protocol Indirect-TCP Explicit Bad State Notification (EBSN) TCP HACK TCP Westwood Mobile TCP (M-TCP) Delayed DUPACK ’ S CSDP Wireless TCP (W-TCP) Wireless Explicit Congestion Notification (WECN) FEC/ARQ protocols

5. Classification of Schemes End-to-End protocols loss recovery handled by sender make the sender realize some losses are due to bit-error, not congestion Link-layer solutions hide link-related losses from sender since the problem is local, solve it locally (local retransmissions) TCP sender may not be fully shielded Split-connection approaches Split each TCP connection into two separate TCP connections at the base station isolate wired network from wireless network

6. Link-layer Solutions Wireless link Wired Internet Mobile host Fixed host BS TCP LL Standard TCP connection Main techniques: Forward Error Correction (FEC) Automatic Repeat Request (ARQ)

7. Link-layer Solutions QoS / Energy trade-off : Applying FEC coding to all of the packets allows high data transfer reliability but increases bandwidth and energy consumption (reduced battery life) When channel conditions are good, ARQ alone obtains better performance using fewer resources !  Hybrid FEC/ARQ solutions

8. Our approach A Link-Layer TCP-aware scheme Combination of FEC and ARQ: FEC is applied only to TCP packets of “ critical ” importance to QoS Convenient trade-off between energy saving and performance End-to-End semantics is maintained

9. Key Observations “ Critical ” TCP packets: packets whose loss would force the sender to wait for a timeout (and reduce the window to one) Timeouts severely affect the completion time of TCP connections Since the radio channel is not congested, timeouts due to channel errors are useless and should be completely avoided !

10. “ Critical ” TCP packets The first 3 segments of a flow : The transmission window is too small to allow the Fast Retransmit mechanism The initial RTO is usually very large The last 3 segments of a flows : There are not enough duplicate ACKs to trigger the Fast Retransmit Retrasmitted segments (by the TCP sender) : If they get lost again, a timeout occurs and RTO is backed-off

11. Simulation scenario 384 Kbps Wired Internet 30 ms delay 10 Mbps TCP receiver TCP NewReno sender Segment size = 1000 bytes BS LL layer implemented into ns-2 simulator according to 3GPP specifications Each TCP segment is divided into 3 LL data units (not protected) or 6 LL data units (protected using rate ½ FEC coding) channel error model: 2-states Markov chain accounting for fading process (see Zorzi, Rao)

12. Qos performance – 10 packets flows 0.0 0.5 1.0 1.5 2.0 123456789101112131415 Average completion time [s] Fading margin, F [dB] no fec selective total Bad radio channel Good radio channel

13. Energy consumption – 10 packets flows Bad radio channel Good radio channel

14. Conclusions … Link-Layer approach to improve TCP over wireless: the best thing to do is a mixture of FEC/ARQ The optimal combination of FEC/ARQ is difficult, because it depends on the channel conditions and on the desired Energy/QoS trade-off We suggest a mechanism based on selective protection of “ critical ” packets, which is particularly effective for short-lived TCP flows (vast majority of Internet flows)

16. Qos performance – 50 packets flows Bad radio channel Good radio channel 0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 123456789101112131415 Average completion time [s] Fading margin, F [dB] no fec selective total

17. Energy consumption – 50 packets flows Bad radio channel Good radio channel 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0 10.0 123456789101112131415 Av. # data units sent per segment Fading margin, F [dB] no fec selective total