Traffic Sensitive Active Queue Management

1
Traffic Sensitive Active Queue Management

• - Mark Claypool, Robert Kinicki, Abhishek Kumar
• Dept. of Computer Science
• Worcester Polytechnic Institute
• Presenter Ashish Samant

2
Introduction

• Internet was not designed to support application
  based quality of service (QoS).
• Best effort model
• No performance guarantees
• Active Queue Management (AQM) helps deal with
  congestion at routers, but is not enough.
• No per application QoS
• Not sensitive to delay/throughput needs
• Our goal, to add per application based QoS to
  AQM, over current best effort Internet.

3
Introduction

• Spectrum of QoS Requirements of Applications

Streaming Video
Interactive Video
Throughput Sensitivity
File Transfer
Streaming Audio
Web-browsing
Interactive Audio
Gaming
Electronic Mail
Delay Sensitivity
4
Introduction

• Problems with previous approaches to AQM
• Static classification
• Per flow state maintenance
• Pricing, policing overhead
• Features of Traffic Sensitive QoS (TSQ)
• Allows applications to indicate delay/throughput
  sensitivity at packet level.
• Can be deployed over existing AQM schemes.
• No significant addition to overhead at routers.

5
Outline

• Introduction
• TSQ Mechanism
• Application Quality Metrics
• Experiments and Results
• Future Work

6
TSQ Mechanism
3
2
1
4
7
TSQ Mechanism
AQM
10 Mbps
q

(hint)

q
TSQ
q
p


5 Mbps
p
Packet queue
10 Mbps
8
TSQ Mechanism

• On receiving each packet, router calculates a
  weight
• w (d td ) / 2N ta
• d delay limit
• td drain time
• N no. of bits used to represent delay hints
• ta arrival time
• Weight of packet determines its position in
  router queue.
• Lower delay hint leads to lower weight.
• Time of arrival prevents starvation.

9
TSQ Mechanism

• TSQ uses cut-in-line scheme to insert packets
  with high delay sensitivity (higher weights)
  towards the front of the queue.
• Packets from throughput sensitive application are
  delegated to the back of the queue
• Packets that cut-in-line are dropped with a
  higher probability to ensure fairness.
• Thus, advantage of labeling packets with high
  delay hints is neutralized with higher drop
  rates.

10
TSQ Mechanism

• The underlying AQM has a drop probability (p)
  that is applied uniformly to all packets.
• Delay sensitive packets receive higher drop
  probability
• p (l q)2 p / (l q)2
• l one way delay
• q instantaneous queue position
• q new queue position
• p drop probability calculated by underlying AQM
• Packets that cut-in-line more will have a
  higher drop probability.

11
Outline

• Introduction
• TSQ Mechanism
• Application Quality Metrics
• Experiments and Results
• Conclusion and Future Work

12
Application Quality Metrics

• Based on previous work, we measure application
  quality as minimum of its delay quality (Qd) and
  throughput quality (Qt)
• Q(d,t) min(Qd, Qt) 0 Q(d,t)
• Higher value of Qd indicates the application is
  more sensitive to delay and vice versa.
• Application quality is normalized between 0 and 1
  
• - 1 indicates highest quality and 0 means no
  quality
• at all.

13
Application Quality Metrics
Interactive Audio Delay Quality Refs
Act02IKK93
Excellent Quality
Good Quality
Excellent Quality
Bad Quality
Good Quality
Bad Quality
14
Application Quality Metrics
Interactive Audio Throughput Quality RefsCor98
15
Outline

• Introduction
• TSQ Mechanism
• Application Quality Metrics
• Experiments and Results
• Conclusion and Future Work

16
Experimental Setup
Network Topology
S1
D1
50 Mbps, 50 ms
S2
50 Mbps, 50 ms
D2
R1
R2
B Mbps
SN-1
DN-1
DN
SN
17
Experimental Setup

• PI parameters a 0.00001822, b 0.00001816, w
  170 Hz, qref 200 packets, qmax 800 packets.
• Average packet size 1000 bytes.
• All experiments run for 100 seconds
• TSQ parameters l 40 ms. This is one-way delay
  constant parameter.

18
Experiment Interactive Audio

• Experiment 1 Interactive Audio
• Bottleneck link bandwidth 15 Mbps
• 100 sources and 100 destinations. One way
  propagation delay 150 ms
• 99 TCP based file transfer flows using delay hint
  16
• 1 TCP friendly CBR source sending at 128 Kbps,
  with varying delay hints.

19
Analysis Interactive Audio
Analysis - Interactive Audio ( Delay )

• Low median queuing delay for lower delay hint.
• Less variation in queuing delay at lower delay
  hints.
• Delay Quality increases as delay hints decrease.

20
Analysis Interactive Audio

• Throughput measured every RTT (300 ms).
• Median throughput low for lower delay hints.

21
Analysis Interactive Audio
Overall Quality

• Overall quality is minimum of delay and
  throughput quality.
• Maximum quality occurs when delay hint is 6.

22
Conclusions

• TSQ provides a per-packet QoS to Internet
  applications.
• It is a best-effort service without any
  guarantees.
• Trade-off between throughput and delay is
  maintained by adjusting queue position and drop
  probability.
• Does not require complex modifications at the
  router, over those needed for the AQM.

23
Future Work

• Derive quality metrics for other applications
  like network games, instant messaging, peer to
  peer.
• Develop applications to dynamically change their
  delay hints.
• Investigate optimum number of bits to be used for
  delay hints.
• Apply TSQ to other domains, for e.g. wireless.

24

• Questions or Comments ?
• Thank you !