12
WebTP: Protocol Design Issues Jeng Lung & Yogesh Bhumralkar

WebTP: Protocol Design Issues

  • Upload
    zhen

  • View
    26

  • Download
    1

Embed Size (px)

DESCRIPTION

WebTP: Protocol Design Issues. Jeng Lung & Yogesh Bhumralkar. Introduction. Following key issues related to the design and testing of the protocol: Congestion Window Control Retransmission Timeout Scheme Performance under Network Jitter. Congestion Control. Background - PowerPoint PPT Presentation

Citation preview

Page 1: WebTP: Protocol Design Issues

WebTP: Protocol Design Issues

Jeng Lung & Yogesh Bhumralkar

Page 2: WebTP: Protocol Design Issues

Introduction

• Following key issues related to the design and testing of the protocol:– Congestion Window Control– Retransmission Timeout Scheme– Performance under Network Jitter

Page 3: WebTP: Protocol Design Issues

Congestion Control

• Background– WebTP uses TCP-style congestion control– 2 Phases:

• Slow Start

• Congestion Avoidance: Additive Increase/Multiplicative Decrease

– TCP biased against long connections; therefore, WebTP faces same dilemma.

Page 4: WebTP: Protocol Design Issues

Improving Fairness

)(*

)(

2

1

2

1

n

i

n

i

i

i

bn

bFairness

• Maximize fairness by modifying Additive Increase

• Instead of increasing cwnd by 1/cwnd, increase it by K=c*rtt*rtt. This makes it more fair

• Problem: Find optimal K for fairness

• K is topology dependent but still want to ensure that the scheme works for WebTP.

Page 5: WebTP: Protocol Design Issues

Network Topology

Page 6: WebTP: Protocol Design Issues

2 Node Cwnd Fairness Study (TCP & WebTP Same RTT)

0

0.5

1

1.5

1 3 5 7 9 11 13 15 17 19

K

Fai

rnes

s

4 Node Cwnd Fairness Study (WebTP has 10 times the RTT of TCP)

00.20.40.60.8

1

1 2 3 4 5 6 7 8 9 10 11

K

Fa

irn

es

s

Page 7: WebTP: Protocol Design Issues

Retransmission Scheme

• Current timeout set to:

• M is the receiver’s tolerance to jitter.

• Examine the effects of changing M on the number of dropped and duplicate packets.

1

M

Page 8: WebTP: Protocol Design Issues

M vs. Drop/Dup Packets

0

500

1000

1500

1 2 3 4 5 6 7 8 9 10 11 12 13 14

M

# P

acke

ts

Drop Packets Duplicate Packets

Page 9: WebTP: Protocol Design Issues

Jitter Study

• Any network has considerable random delays.

• The time between when the packet is transmitted to when it reaches the receiver varies a great deal - this phenomenon is called jitter.

• Simulated jitter by introducing random delay to each packet on the sender side.

Page 10: WebTP: Protocol Design Issues

WebTP with no jitter

0

5000

10000

15000

1 2 3 4

M

Se

nt

Pa

ck

ets

WebTP with [0, 3ms] jitter

0

5000

10000

15000

1 2 3 4 5 6

M

Se

nt

Pa

ck

ets

Page 11: WebTP: Protocol Design Issues

WebTP with no jitter

0

500

1000

1500

2000

1 2 3 4

M

Du

plic

ate

Pa

ck

ets

WebTP with [0, 3ms] jitter

0

500

1000

1500

2000

1 2 3 4 5 6

M

Du

plic

ate

Pa

ck

ets

Page 12: WebTP: Protocol Design Issues

Conclusion

• Congestion Window Scheme: Optimal K depends on network topology.

• Retransmission Scheme: Dropped packets increase with M whereas the number of duplicate packets goes down at higher M.

• Network Jitter: Higher jitter implies that a higher M is required to handle the delays that are introduced.