Improving TCP over Wireless Environments using Cross Layer Feedback

1
Improving TCP over Wireless EnvironmentsusingCr
oss Layer Feedback

• Vijay T. Raisinghani (Tata Infotech KReSIT, IIT
  Bombay)
• Ajay Kr. Singh (CSE, IIT Bombay)
• Prof. Sridhar Iyer (KReSIT, IIT Bombay)
• rvijay_at_it, aksingh_at_cse, sri_at_it.iitb.ac.in

http//www.cse.iitb.ac.in/tirg
http//www.it.iitb.ac.in
2
Typical Scenario

• FH Fixed Host
• MH Mobile Host
• BS Base Station (gateway)

3
Mobile Wireless Networks

• MWN characteristics
• High bit error rate of wireless channel
• Mobility induced disconnections
• TCP characteristics
• End-to-end reliability
• Dynamically adapts sending rate based on network
  congestion (congestion window) and receiver
  buffer (advertised window)
• Assumes all packet losses due to congestion

4
TCP over wireless networks

• On packet loss
• TCP assumes network congestion
• reduces throughput
• In wireless networks
• many packet losses are due to bit errors,
  mobility
• TCPs congestion assumption fails
• reduction in send window inappropriate

5
Optimizing TCP for MWN

• FH Fixed Host
• MH Mobile Host
• BS Base Station (gateway)

6
Optimizing TCP for MWN

• Several approaches that focus on mitigating
• Adverse effect of wireless channel
• Mobility induced disconnections
• Any approach involves one or more of
• Modification to TCP stack at the Fixed Host (FH)
• Per-connection support at the Base Station (BS)
• Modification to TCP stack at Mobile Host (MH)
• Typically assume TCP sender at the FH

7
Our focus

• Modification to TCP stack at MH only
• Optimizing TCP to mitigate effect of mobility
  induced disconnections
• Focus on two-way data transfer
• TCP sender at Mobile Host (MH) and/or
• TCP sender at Fixed Host (FH)
• Approach cross-layer feedback

8
Cross Layer Feedback Motivation

• Layered protocol stack design is good software
  engineering
• Strictly layered stacks do not perform well over
  wireless networks
• network conditions are highly variable
• mobile nodes are resource poor
• Several assumptions from fixed wired networks do
  not hold for wireless

9
Observations

• Cross layer information can help improve
  performance over wireless networks
• Upper to lower layers
• TCP timer information
• application QoS requirements
• user feedback
• Lower to upper layers
• link characteristics
• network connectivity status

10
Cross Layer Feedback Some Existing Approaches

• TCP QoS information to adapt link layer
  retransmissions (Chiasserini, Meo. IEEE Globecom,
  2001)
• TCP fast retransmit (Caceres, Iftode. IEEE JSAC
  95)
• Layer2 info to MobileIP for IP handoff (Wu, et
  al. MONET, 2001 )
• No approach employs user inputs

11
User feedback

• User feedback to protocol stack could be
• Impending disconnection information
• Dynamic changes in application priorities
• For example In view of impending disconnection,
  an ongoing FTP may become more important than an
  ongoing video conference contrary to default
  system priorities
• System can avoid performance degradation
• mapping user input to protocol specific actions

12
Our approach for TCP optimization

• Receiver Window Control
• Based on user feedback idea
• User indicates application priorities
• System maps priorities to TCP receiver window
  control
• Retransmit Timer Control
• Based on network layer feedback
• Network indicates connectivity status
• System appropriately modifies TCP retransmission
  mechanism

13
BackgroundTCP receiver window

• Reflects receivers available buffer through
  advertised window (awnd) in ACKs
• Optimum awnd bandwidthdelay (bdp) to fill pipe
  and maximize sender throughput
• awnd lt bdp decreases sender throughput
• Each application on MH may require different
  awnd, according to bdp

14
Receiver Window Control (RWC)

• Exploits idea Sender throughput decreases as
  awnd lt bdp
• Higher awnd for high priority applications
• Restrict awnd for low priority applications
• Assume ?awnd is a fixed resource
• (Re)distribute ?awnd according to priority

15
RWC details
Advertised window before user feedback

• Following condition holds

Number of Applications
Advertised window after user feedback
Bandwidth to MH on bottleneck link (constant)
rtt of all applications on MH (constant)
16
RWC details

• User changes priority of application k from xi to
  xi
• Change awnd for all applications

17
RWC example

• A 30, x1 1, x2 1, x3 1
• Thus, awnd 10 for all applications
• After feedback, x12 (x2x31)
• awnd according to relative application priority

18
RWC simulations

• Implemented in ns2

• B 2 Mbps
• R 64ms

19
RWC results

• A32 packets
• x1x21
• awnd1awnd216

20
RWC results

• A32
• x12 x21
• awnd211 awnd121
• 31 change

21
RWC results

• 0.1 loss
• No RWC
• Thruput variations due to random losses

• A32 packets
• x1x21
• awnd1awnd216

22
RWC results

• 0.1 loss
• Thruput still increases for higher priority
  application

23
Our approach

• Receiver Window Control
• Based on user feedback idea
• User indicates application priorities
• System maps priorities to TCP receiver window
  control
• Retransmit Timer Control
• Based on network layer feedback
• Network indicates connectivity status
• System appropriately modifies TCP retransmission
  mechanism

24
Send Window open Disconn
25
MH to FH data transfer

• Upon disconnection event
• Set ConnectionStatus Disconnected
• Set DisconnectionOccurred True
• If Sending Window is Open
• Stop sending
• Cancel RTX (retransmit timer)
• If Sending Window is Closed
• Wait for RTO (expiry of RTX timer)

26
Send window closed Disconn
Case 2 State Send window closed and waiting for
acks at disconnection event. Action retransmit
the last unacked packet, with ssthresh set to
window reached till disconnection, at
reconnection.
27
MH to FH data transfer (contd)

• Upon connection event
• Set ConnectionStatus Connected
• If Sending Window is Open
• Set RTX
• Resume transmission
• If Sending Window is Closed
• Wait for RTO

28
Send window closed ack wait
29
MH to FH data transfer (contd)

• At RTO (expiry of retransmit timer)
• If ConnectionStatus Disconnected
• Set ssthresh cwnd
• Set cwnd 1
• If ConnectionStatus Connected and
  DisconnectionOccurred True
• Retransmit unACKed packets without modifying
  ssthresh or cwnd
• Set DisconnectionOccurred False

30
MH actions TCP sender at FH

• MH delays the ACK of last two bytes (d msec)
• Upon disconnection event
• Update network connectivity flags
• Upon connection event
• Send ZWA (Zero Window Advertised) with ACK for
  second last byte
• Send FWA (Full Window Advertised) with ACK for
  last byte
• FH resumes transmission without reducing cwnd

31
Simulations

• Implemented in ns-2
• Augment mobile IP for connectivity information

32
MH to FH data transfer (RTT 5ms)
33
MH to FH data transfer(RTT700ms)
34
FH to MH data transfer(RTT 5ms)
35
Conclusion

• Our approach to optimizing TCP in MWN
• Mitigates effect of mobility induced
  disconnections
• Requires modification to TCP stack at MH only
• Uses lower layer feedback for connectivity status
• Uses upper layer feedback for dynamic priorities
• Novelty User in the loop

36
Future Work

• Explore other aspects of user feedback
• Identify other scenarios where cross layer
  feedback may be useful
• ARP in mobile ad hoc networks
• Focus on devising new mechanisms for implementing
  cross layer feedback
• Vertical signaling stack

37
THANKS!
38
Summary

• Cross Layer Feedback is useful in wireless
  networks
• New approach User feedback can be exploited
• Receiver window control one method of exploiting
  user feedback
• RWC preliminary simulations show promising results

39
Future Work

• Focus on exploring new mechanisms for cross layer
  feedback
• Explore other aspects of user feedback

40
RWC for varying rtt
41
RWC fairness

• Packet arrivals Poisson for each flow
• Properties

?1
?1
?
?
?2
?2
...
...
Bottleneck link
?n
?n
42
RWC fairness (contd.)
?1
?1
?2
?2
?RWC
?RWC
?3
?3
?others
?others