When TCP Friendliness Becomes Harmful

1
When TCP Friendliness Becomes Harmful

• Amit Mondal
• Aleksandar Kuzmanovic
• Northwestern University

http//networks.cs.northwestern.edu/
2
Introduction

• Internet was designed for throughput
• TCP probes for available b/w even if it causes
  packet losses
• Interactive applications suffer!
• Telnet, ssh, games, chat all use TCP
• Telnet delay from typed character until echo
• Includes transmission, propagation and queuing
  delay
• It loss, TCP retransmits

3
Upgrading mice to elephants
data packets
strict priority
TCP-fair rate
dummy packets
Padding misbehavior
4
Outline

• Source of the problem
• Related work
• Implication
• Padding-induced response time gain
• Sustainable countermeasures
• Conclusion

5
Outline

• Source of the problem
• Related work
• Implication
• Padding-induced response time gain
• Sustainable countermeasures
• Conclusion

6
Source of the problem

• How TCP detects packet loss?
• Retransmission Timeout (RTO) (1 second)
• Triple duplicate acknowledgment/Fast retransmit
  ( 10-100 ms)
• How does it affect interactive applications?
• Burst size ( 1-2 packets)
• Inter-burst time ( seconds)

Packet loss is detected by timeout, which
degrades the response time in orders of magnitude
7
A deeper look
RTO
Incentive for misbehavior!
8
Outline

• Source of the problem
• Related work
• Implication
• Padding-induced response time gain
• Sustainable countermeasures
• Conclusion

9
Related Work

• Packet marking differential dropping
• Guo and Matta 01
• Service differentiation
• Noureddine and Tobagi 02
• Differential Congestion Notification
• Le et al. 04
• Explicit congestion notification
• Floyd 94
• TCP smart framing
• Mellia et al. 05

10
Outline

• Source of the problem
• Related work
• Implication
• Padding-induced response time gain
• Sustainable countermeasures
• Conclusion

11
Implication
Packet switched gt Circuit switched
12
Outline

• Source of the problem
• Related work
• Implication
• Padding-induced response time gain
• Sustainable countermeasures
• Conclusion

13
Quantifying padding induced-gain
P packet loss ratio Q prob. packet loss is
detected timeout
Correlated packet loss (FIFO)
Random packet loss (RED)
14
Modeling
Upgrading interactive TCP flows to
fully backlogged flows always pays-off
15
Simulations
1) Queuing delay plays a role
2) Consecutive packet losses can affect overall
gain achieved
16
Outline

• Source of the problem
• Related work
• Implication
• Padding-induced response time gain
• Sustainable countermeasures
• Conclusion

17
Sustainable countermeasures (1)

• Differentiated minRTO
• Application-limited flows use reduced value for
  minRTO parameter
• Short-term padding with dummy packets
• Enable that a packet loss is detected via fast
  retransmit mechanism
• Actual packet followed by three tiny dummy
  packets.
• A diversity approach
• TCP sends k (kgt1, k is a small integer) copies of
  the packet without violating congestion control
  mechanism
• In reality k2 is sufficient

18
Sustainable countermeasures (2)
Differentiated minRTO does not work
Short-term padding and diversity approaches
outperform fully-backlogged approach
19
Sustainable countermeasures (3)
RED
FIFO
Short-term padding and diversity approaches
outperform fully-backlogged approach
20
Overhead and Sustainability
Issue What happens when everyone adopts our
approach?
Short-term padding and diversity approaches
provide significantly friendlier environment
21
Conclusion

• TCPs loss recovery mechanism degrades response
  time of interactive applications relative to long
  flows during congestion.
• Incentive for padding misbehaviour and seriously
  degrades overall networks performance.
• Explored simple, sustainable and easily
  deployable solutions.

22
Questions?