Service Differentiation at Transport
Layer via TCP Westwood LowPriority (TCPW-LP)
H. Shimonishi, M.Y. Sanadidi and M. Geria
System Platforms Research Laboratories, NEC Corporation
UCLA Computer Science Department
IEEE Symp on Computers & Communications (ISCC), 2004
1
Outline
 Introduction
 TCP Westwood (TCPW)
 TCP Westwood Low Priority (TCPW-LP)
 Performance Evaluation


Coexistence with foreground traffic
Comparison of TCPW-LP and TCP-LP
 Conclusion
2
Introduction
 TCP Westwood Low-Priority (TCPW-LP)


An end-to-end “foreground/background”
priority scheme
Objectives



Non-intrusive to coexisting foreground traffic
Capable of fully utilizing the unused bandwidth
Capable of fairly sharing with other low-priority
flows
3
Introduction
 Application


Web objects pre-fetching (cache)
Large bulk transfers, e.g. FTP
4
Introduction
 Related Works

DiffServ (proposed by IETF)


Support from the network router is required
End-to-end schemes (TCP-LP and TCP-Nice)


Unused bandwidth cannot be fully utilized
Pre-set queuing threshold is required
5
Background - TCPW
 TCPW – a sender-side only modification
 Reaction to packet losses

Duplicate ACKs

Reno
 CWIN = CWIN/2

Westwood
 CWIN = (BWE * RTTmin)

Timeout expiration

Reno and Westwood
 CWIN = 1
6
Background - TCPW
 BWE – Bandwidth Estimation

Estimated from the rate of ACK

b = segment size / (ACKtime - lastACKtime)
 segment size = average of last n received segment

BWE = αBWE + (1- α)*b
 smoothing operator α=0.8
7
TCPW-LP
 Early Window Reduction (EWR)

Congestion window reduction scheme
 Dynamic Threshold Adjustment

Foreground Traffic Ratio, r
8
Early Window Reduction (EWR)
 Limit the backlog over the path
 Virtual queue length = CWIN – BWE*RTTmin


CWIN = amount of outstanding packets in the
path
BWE*RTTmin = amount of packets in the virtual
pipe
9
Early Window Reduction (EWR)
 The virtual queue length exceeds a threshold

CWIN = BWE*RTTmin – BWE*Da


Da – the average queuing delay
BWE*Da – the packets backlogged at the
bottleneck
10
Dynamic Threshold Adjustment
 Foreground Traffic Ratio (FTR), r


Ratio of Temporal Minimum Queuing Delay to
Average Queuing Delay
When all queued packets belong to
foreground traffic

r approaches 1
 only background flows
 minimum queuing delay is small due to EWR
 average queuing delay grows according to the
backlog threshold
11
Dynamic Threshold Adjustment
 Dynamic Threshold, Qth = M(1-r)

M = 3 (upper bound on backlogged packets)
 FTR, r = Dm /(Da+δ)


Dm = αDm + (1-α) Dmin
Da = αDa + (1-α) Davg


α= 3/4
δ= 3x10-6/(RTT-RTTmin), ensuring non-zero
delay in the calculation of r
12
Performance Evaluation
 Simulation Topology


End-to-end round trip propagation delay =
74ms
FIFO queuing with drop tail discipline
13
Coexistence with foreground traffic
 Throughput
14
Coexistence with foreground traffic
 Congestion Window Behavior
15
Coexistence with foreground traffic
 Completion time evaluation using FTP traffic
16
Coexistence with foreground traffic
 Effect of packet losses
17
Comparison of
TCPW-LP and TCP-LP
 Throughput


20 identical flows
TCP-LP flows utilize only 68% of the link
18
Comparison of
TCPW-LP and TCP-LP
 Effect of packet losses
19
Comparison of
TCPW-LP and TCP-LP
 Coexistence with UDP traffic



On-off UDP traffic
Available Bandwidth = 3.3Mbps(On),
10Mbps(Off)
Average available bandwidth = 6.7Mbps
20
Comments
 Some Questions
 TCP-LP, one-way delay?
 Analytical study of Foreground Traffic Ratio?
 Packet loss improvement? TCP Westwood?
 Insight
 No bandwidth guarantee in both TCPW-LP
and TCP-LP
 Protocol between ordinary TCP and TCPWLP/TCP-LP
 Receiver-side only modification scheme
21
Conclusion
 TCPW-LP – an end-to-end scheme to realize
two-class service prioritization
 Dynamically adjusting the queuing threshold
 Evaluation of its performance by simulation
 Comparison of TCPW-LP and TCP-LP
22