ppt

WHITE – Achieving Fair Bandwidth
Allocation with Priority Dropping
Based on Round Trip Time
Name : Choong-Soo Lee
Advisors : Mark Claypool, Robert Kinicki
Reader : Craig Wills
Date: March 25, 2002
Outline

Introduction




Related Work
Approach
Evaluation
Conclusion
Introduction

Current internet uses routers with droptail
queue management


Droptail introduces the problem of global
synchronization
There are many active queue managements
proposed but most of them are concerned
with overall throughput and delay but not
with fairness

Flows are not homogeneous but heterogeneous

Robust flows vs. Fragile flows
Related Work




Random Early Detection (RED)
Flow RED (FRED)
Core-Stateless Fair Queuing (CSFQ)
Deficit Round-Robin (DRR)
RED [FJ93]
Based on average queue size

Maxth
Minth
1
max_p
0
minth
maxth
queue
size
FRED [LM97]


Modification to RED
Maintains per-flow state information
CSFQ [SSZ98]


Rate-based Active Queue Management
Distinguishes between edge and core
routers



Edge routers label packets
Core routers use these labels to treat packets
fairly
Estimates fair share and uses it to drop
packets
DRR


Implementation of Fair Queuing
Maintains per-flow state information
Overview
Per Flow
Per Packet
No Per Packet

DRR
No Per Flow
FRED
WHITE
CSFQ
RED
Goals


Achieve fair allocation close to Fair Queuing and comparable or
better than RED, FRED and CSFQ under most scenarios.
Reduce complexity by not having to maintain per flow state
Outline


Introduction
Approach





Round Trip Time at the Edge
Average Round Trip Time at the Router
Drop Probability Based on Round Trip Times
Evaluation
Conclusion
Approach

Modification to RED


Adjusts max_p per packet
Supports both dropping and marking of packets






Dropping vs. Marking
Dropping WHITE : Chardonnay
Marking WHITE : Chablis
Round Trip Time at the Edge
Average Round Trip Time at the Router
Drop Probability Based on Round Trip Times
Round Trip Time at the Edge

Edge Hint




Packets get labeled with additional information
We want the lowest RTT as our hint
Modification to TCP-Reno with TCP-Vegas
RTT Computation
4-17 bits in the IP header available for
additional information if no fragmentation
[SZ99]
Average Round Trip Time at the
Router

Now that we have the RTT edge hint, RTTs
are exponentially weighted (Raverage) at the
router
R average  1  w RTT  R average   w RTT  p.RTT

Due to high fluctuation of Raverage, we use
extra steps to compute stabilized value of
RTT (Rformula)

How long it has been out of 12.5ms
Drop Probability Based on Round
Trip Time


Now, we want to use RTT edge hint and average
RTT at the router to compute drop probability
TCP-Friendly Formula [PFK98]
s
T
R

2p
3p
 t RTO
p 1  32p 2 
3
8
Simplify
s
T
cRp a
T1 = T2
Drop Probability Based on Round
Trip Time
T1  T2
s
s

a
cR 1 p1
cR 2 p 2 a
p2
a
R1
p
R2
a
1
 R1 
 p 2  p1 

 R2 
p robust
p fragile
1
a
 R formula 
 p base 

R
 robust 

 R formula
 p base 
R
 fragile




Drop Probability Based on Round
Trip Time
1.2
p
 p  p
3
7
3.5
 cp
p
1.58
 p  p
3
7
 c  p0.71
3
1
2.5
0.8
2
0.6
1.5
0.4
1
0.2
0.5
0
0
0
sqrt(p)
0.1
sqrt(p)^3
0.2
0.3
Drop Probability
sqrt(p)^7
0.4
Sum
0.5
Power (Sum)
0.5
0.6
sqrt(p)
sqrt(p)^3
0.7
0.8
Drop Probability
sqrt(p)^7
0.9
Sum
1
Power (Sum)
Drop Probability Based on Round
Trip Time



For Chardonnay, 0.71 corresponds to 
robust) and 1.58 to  fragile).
For Chablis, 1.58 corresponds to both 
robust) and  fragile).
However, simulation results showed that
values of (0.65, 1.4) worked the best for
Chardonnay and (1.6, 1.4) for Chablis.
WHITE Algorithm
qave
1
max_p
0
minth
maxth
robust flow
fragile flow
queue
size
Outline



Introduction
Approach
Evaluation




Setup
Experiments
Chardonnay vs. Chablis
Conclusion
Setup



Network Simulator 2 (NS-2) was used to
run all the simulations.
Modification to source code to include RTT
edge hints and to implement WHITE.
We ran 6 experiments with RED, FRED,
CSFQ, DRR, Chardonnay and Chablis
Setup
N0
5 Mbps
Queue Size: 120
10 Mbps, 5ms
R
N1
N2
N29
RED/FRED
minth: 10
maxth: 30
wq:
0.0008
max_p: 0.1
WHITE
(Chardonnay,
Chablis)
minth: 10
maxth: 30
W q:
0.0008
max_p: 0.1
:
0.65, 1.6
:
1.4, 1.4
D
CSFQ
K:
K :
Kc:
100ms
100ms
100ms
Experiments

Uniformly Distributed Latencies (Exp1)




Round trip latencies from sources were 20ms,
30ms, 40ms, … , 310ms.
Balanced Clustered Latencies (Exp2)
Unbalanced Latencies (Exp3, Exp4)
Dynamic Latencies (Exp5, Exp6)
Uniformly Distributed Latencies
Uniformly Distributed Latencies
Uniformly Distributed Latencies
Uniformly Distributed Latencies
Uniformly Distributed Latencies
Uniformly Distributed Latencies
Experiments



Uniformly Distributed Latencies (Exp1)
Balanced Clustered Latencies (Exp2)
Unbalanced Latencies



1 flow with 20ms round trip latency and 29
flows with 200ms round trip latency (Exp3)
1 flow with 200ms round trip latency and 29
flows with 20ms round trip latency (Exp4)
Dynamic Latencies (Exp5, Exp6)
Unbalanced Latencies:
1 Robust vs. 29 Fragile
Unbalanced Latencies:
1 Robust vs. 29 Fragile
Unbalanced Latencies:
1 Robust vs. 29 Fragile
Unbalanced Latencies:
1 Robust vs. 29 Fragile
Unbalanced Latencies:
1 Robust vs. 29 Fragile
Unbalanced Latencies:
1 Robust vs. 29 Fragile
Unbalanced Latencies:
1 Fragile vs. 29 Robust
Unbalanced Latencies:
1 Fragile vs. 29 Robust
Unbalanced Latencies:
1 Fragile vs. 29 Robust
Unbalanced Latencies:
1 Fragile vs. 29 Robust
Unbalanced Latencies:
1 Fragile vs. 29 Robust
Unbalanced Latencies:
1 Fragile vs. 29 Robust
Experiments




Uniformly Distributed Latencies (Exp1)
Balanced Clustered Latencies (Exp2)
Unbalanced Latencies (Exp3, Exp4)
Dynamic Latencies

10 flows with 50ms round trip latency, 10 flows
with 100ms round trip latency and 10 flows
with 200ms round trip latency (Exp6)
Dynamic Latencies
Robust
Average
Fragile
0s
30s
A
60s
B
90s
C
120s
D
Dynamic Latencies
Dynamic Latencies
Dynamic Latencies
Dynamic Latencies
Dynamic Latencies
Dynamic Latencies
Overall Comparison
1.00
Jain's Fairness
0.95
0.90
0.85
0.80
0.75
0.70
1
3
RED
4
FRED
6A
Experiment
CSFQ
DRR
6B
Chardonnay
6C
Chablis
6D
Jain's Fairness Index
Chardonnay (Dropping) vs.
Chablis (Marking)
1
0.98
0.96
0.94
0.92
0.9
0.88
0.86
0.84
0.82
1
2
3
4
5A
5B 5C
5D 6A
Experiment
Chardonnay
Chablis
6B 6C
6D
Chardonnay (Dropping) vs.
Chablis (Marking)
Experiment
Chardonnay
Drop (%)
Chablis
Goodput
(Mbps)
Drop (%)
Goodput
(Mbps)
1
1.80
9.59
0.000
9.65
2
2.70
9.91
0.000
9.98
3
1.46
9.67
0.000
9.78
4
3.59
9.96
0.007
9.96
5
2.56
9.76
0.002
9.85
6
2.49
9.69
0.003
9.82
Outline




Introduction
Approach
Evaluation
Conclusion

Future Work
Conclusion

Performance of Chardonnay and Chablis is
better than RED, FRED and CSFQ and
comparable to DRR


RTT edge hints can be used to approximate
DRR’s performance without the complexity of
maintaining per-flow state information
Marking performed better


Less drops
Better goodput
Future Work


Current version of WHITE does not support
any non-responsive flows such as UDP
flows
Adaptive mechanism is necessary to support
much more flows than those in simulations