Understanding VoIP from Backbone Measurements

1
Understanding VoIP from Backbone Measurements

• Marco Mellia, Dario Rossi Robert Birke, and
  Michele Petracca
• INFOCOM 07, Anchorage, Alaska, USA
• Young J. Won
• Oct. 15, 2007

2
Outline

• Introduction
• Measurement methodology
• The FastWeb network
• Measurement results
• Conclusion

3
Introduction

• VoIP has long been indicated as the technology
  that will trigger convergence.
• Traffic monitoring and characterization have been
  seen as a key methodology to understand
  telecommunication technology and operation
• This paper presents the first extended set of
  measurement results collected via passive
  monitoring of VoIP traffic

4
Measurement Methodology

• Identification of RTP/RTCP over UDP flows
• Measurement indexes
• Call duration
• Call round trip time
• Flow packet loss probability
• Flow jitter (Inter-Packet-Gap variation)
• Flow equivalent mean opinion score (eMOS)
• A computational model by ITU-T the predicts
  subjective quality of packetized voice.

5
Why?

• Why we are interested in this?
• Curiosity
• Extracting basic parameter values for OPNET and
  later modeling
• Basis for the theorical scenario analysis models

6
Identification of RTP/RTCP over UDP Flows

• Conditions to Check
• The version field must be set to 2
• The payload type field must have an admissible
  value or the same SSCR (Synchronized Source
  Identifier)
• The UDP port gt 1024 or the same payload type

7
The FastWeb Network
8
Measurement Summary

• July 15, 2006 - 10 am to 2 pm (4 hr)
• 240GB packet header
• 150,000 phone calls, RTP and RTCP over UDP
• Voice transport, G.711a Codec
• Two 64kbps streams, 50 packets per sec
• Packetization time is set to 20 ms
• No per-class differentiation
• The maximum values are observed between 10 am to
  2 pm
• More than 1300 simultaneous calls per minute
• Drops in lunch break and in the early afternoon

9
Measurement Results

• Two probe nodes located in a PoP located in
  Turin, and a Gateway node located in Milan
• Tstat was run on the probe to take live traffic
  measurements
• Passive monitoring tool by the Politecnico di
  Torino
• RTP/RTCP over UDP or tunneled TCP

10
HAG Distribution

• Home Access Gateway (HAG)
• Offers Ethernet ports to PCs, VideoBox, Plain Old
  Telephone

11
Number of Phone Calls Tracked
12
User-Centric Measurements (Call Duration)

• User-Centric measurements
• Average phone duration, 106s heavy-tailed
  distribution
• Long (97s), Local (113s)
• My observation Very similar to FLOW Lifetime
  measurement

13
eMOS CDF

• eMOS
• Excellent ( eMOSgt4 )
• Good ( eMOS34 )
• eMOS gt 3.6
• The same quality as traditional PSTN phone calls

14
Network-Centric Measurements (RTT)

• All measurements present RTT values smaller than
  200 ms for more than 97 of calls
• RTT cannot be considered as a major impairment of
  VoIP call quality

15
Network-Centric Measurements (Jitter)

• Jitter is traced smaller than 15ms
• Inter-packet-gap is 20ms

16
Network-Centric Measurements (Packet Loss)

• Average loss probability 2.8
• The bottom picture showed that packet loss
  probability has little correlation with the
  actual network load

17
Conclusion

• This paper presented an extensive measurement
  campaign focusing on VoIP traffic
  characterization
• Use eMOS model to compare the quality of VoIP to
  traditional PSTN phone calls
• In FastWeb, only the packet loss probability
  affected the quality of VoIP