Transport Control Protocol for Wireless Connections

1
Transport Control Protocol for Wireless
Connections

• ElAarag and Bassiouni
• Vehicle Technology Conference 1999

2
I. Introduction

• Technical problems of the wireless mobile
  networks
• Lossy (higher bit error rate)
• Mobile disconnections (e.g. hand-off)
• This paper studies
• The effect of bit error rate and mobile
  disconnection on the performance of different TCP
  implementations (Tahoe, Reno, New Reno, SACK)
• This paper shows
• The performance of TCP is more sensitive to the
  link up period than the error rate and the
  disconnection probability
• SACK outperforms Reno in wired networks, but not
  in the wireless case.

3
II. Simulation Model

• Network Simulator (NS) by Lawrence Berkeley Labs
• Three nodesA fixed host (FH), a base station
  (BS), and a mobile host (MH)
• Good state the wireless link suffers no
  loss exponentially distributed with means
  agBad state the wireless link suffers from
  packet loss exponentially distributed with
  means a b
• MH
• Link-up state exponentially distributed with
  mean link-up time m1
• Link-down state exponentially distributed with
  mean link-down time m2
• Disconnection probability p m2 / (m1m2)

4
TCP Parameters

• Windows size 50
• ssthresh 32
• Packet size 1024
• ACK size 40
• TCP clock granularity 100 ms
• Queue size 50
• Focus on
• loss of packets due to mobile link error and
  mobile disconnections

5
Performance Metrics

• Four parameter metrics
• Throughput
• How much data the user can receive per second
• Goodput
• The ratio between the amount of useful data
  transmitted by the source and the mount of data
  transmitted by the source
• Packet delay
• The average delay incurred by the packet from the
  time it is deposited into the senders windows
  until it is successfully acknowledged
• Transfer time
• The time it takes for the receiver to receive a
  fixed number of packets, e.g. 5000 packets
• Better quality of service means

6
III. Effect of Mobile Disconnection Probability

• Experiment
• on Tahoe, Reno, New-Reno, and SACK TCP
• Two scenarios
• Scenario 1 p 9 (m11.0s, m20.1s)
• Scenario 2 p 28.6 (m15.0s, m22.0s)
• Observation
• TCP Performancedepends on
• disconnectionprobability???
• Link up time!!!?The sender has enough time to
  open up a large windowbefore it disconnects

7
IV. Effect of Link Up Period

• The larger the mean link up time,the better
  the throughout, the good put, and the transfer
  time

8

• The larger the mean link up time,the better
  the throughout, the goodput, and the transfer
  time

9

• The larger the mean link up time,the better
  the throughout, the goodput, and the transfer
  time

10
V. Performance of Tahoe, Reno, New-Reno, and SACK
TCP in Wireless Networks

• Ideal case (no loss)
• Lossy wireless link (Fig.46)
• BER 10-5
• Data (BS ?MH) ag100 ab8
• ACKs (BS ?MH) ag1000 ab3
• m12.0 m20.3

11
(No Transcript)
12
(No Transcript)
13
(No Transcript)
14
(No Transcript)
15
cwnd

• Tahoe has the worst performance?
• the least average cwnd
• the largest number of occurrences of cwnd1
• Reno has the best performance?
• Not the largest average cwnd
• The largest number of timeouts,the most
  sensitive in changing cwnd, thus the most
  adaptive to the network need

16
VI. Conclusion

• The mobile link up period has dominant effect on
  the performance of TCP than the disconnection
  probability of the mobile and even the BER of the
  wireless link.
• Comparing Tahoe, Reno, New-Reno, and SACK TCP in
  the presence of wireless link errors and mobile
  disconnections
• Reno had the most throughput, least average
  delay, and least transfer time.