Multipoint QoS Measurements with QoSMeT Pekka Per

1
Multipoint QoS Measurements with QoSMeTPekka
Perälä, Jarmo Prokkola and Marko
JurvansuuConverging Networks Laboratory -
VTTpekka.perala_at_vtt.fiTel 358 400 989
353P.O. Box 1100 (Street Kaitoväylä 1,
Oulu)FI-90571 Oulu, Finland23.09.2008
2
Contents

• Introduction
• Overview of QoS Measurement Tool (QoSMeT)
• The multipoint measurements with QoSMeT
• Measurement examples
• Conclusions

3
Introduction (1)

• Real-time applications demand high quality from
  the network (low delay, low jitter, low packet
  loss, etc.)
• High available BW does not always ensure QoS
• Many methods reveal the overall performance of
  the network, but it does not necessary reflect
  well on how the individual applications behave.
• ? It is important to find out the QoS metrics of
  a single application
• The passive E2E QoS measurements reveal the QoS
  experienced by an application
• VTT's QoSMeT tool measures one-way QoS metrics
• For e.g. SLA verification
• However, E2E measurement is not able to locate
  the problematic link, it only can indicate that a
  performance problem exists
• ? One solution to the problem is multipoint
  measurement that reveals the performance on per
  network segment basis.

4
Introduction (2)

• Multipoint QoS measurements are continuation to
  the development work of QoSMeT (ITEA Easy
  wireless Celtic Easy Wireless 2)
• Objective is to maintain QoS while roaming over
  heterogeneous networks
• The implemented demonstrator integrates
  traditional network analyzer and E2E QoS
  measurement
• QoSMeT and multipoint measurement systems have
  been tested not only in controlled environment,
  but also in public networks (HSPA, public Wi-Fi,
  WiMAX, etc.)
• M. Jurvansuu, J. Prokkola, M. Hanski, P. Perälä,
  HSDPA Performance in Live Networks, In
  proceedings of IEEE International Conference on
  Communications (ICC 2007).
• Our Converging Networks Laboratory (CNL) has been
  the main facility for the tests in addition to
  public networks

5
QoSMeT Overview (1)

• QoSMeT is a passive tool software for measuring
  one-way end-to-end network performance in
  real-time from the applications point of view
• Designed for measuring especially real-time
  applications (e.g., VoIP, video conferencing),
  but other networking applications can be measured
  as well.
• The tool works practically over any kind of
  network as long as IP is supported
• No additional hardware
• One-way delay measurements require GPS time
  stamping (and GPS equipment)
• Win32 platform (Linux is coming)
• J. Prokkola, M. Hanski, M. Jurvansuu, M.
  Immonen, Measuring WCDMA and HSDPA Delay
  Characteristics with QoSMeT, In proceedings of
  IEEE International Conference on Communications
  (ICC 2007).

6
QoSMeT Overview (2)

• A separate TCP connection is used for carrying
  the QoSMeT control data over the network if
  real-time mode is used
• Application is not aware of the measurement
• Records in real-time
• Average QoS statistics
• Accurate per packet statistics
• The most important measured QoS metrics
• Delay
• Jitter (average and absolute)
• Packet loss
• The length of a connection break (for example
  during a vertical handover)
• Traffic statistics
• Pseudo-subjective analysis VoIP MOS (Mean
  Opinion Score) quality

7
QoSMeT Measurement Examples WCDMA and HSPA
Delay and Goodput Performance
J. Prokkola, P. Perälä, M. Hanski and E. Piri
3G/HSPA Performance in Live Networks from the End
User Perspective , submitted to IEEE ICC2009
8
Multipoint Measurement Principle

• Multipoint measurements help to locate the
  performance bottlenecks by revealing the QoS of
  each network segment
• P. Perälä, M. Jurvansuu, J. Prokkola, Combined
  Terminal and Network Measurement System for
  Bottleneck Localization, In proceedings of
  Tridentcom 2008
• The traffic is captured from several (gt 2)
  measurement points (MP), where the traffic is
  captured
• The one-way performance can be split to the
  performance of network segments based on the
  captured data
• ? The segment with the poorest performance can be
  found

According to multipoint measurement principle the
network performance is divided into performances
of network segments
9
Multipoint measurement system

• GPS timestamping enables breaking-down the
  one-way delays
• Integration of QoSMeT and Nethawk M5 analyzer
• Each measurement point captures the data of
  interest (generated by single or multiple
  applications)

Multipoint measurement system applied to UMTS.
10
Public Wi-Fi network multipoint measurements (1)

• Conducted in panOulu, a city-wide WLAN (802.11g)
  network
• Three measurement points (two QoSMeTs and a
  commercial network analyzer)
• One-way delay, access network delay and Internet
  delay were calculated
• Packet loss was found interesting too
• UDP streams of different rates

Figure 9. panOulu measurement setup.
E2E delay (ms) Access network (ms) Internet (ms)
Average delay 14.49 2.85 11.64
Figure 10. One-way delay statistics for VoIP call
(steady state performance)
11
Public Wi-Fi network measurements (2)

• In steady state the Internet delay is dominating
  (Fig. 12)
• After max bandwidth (lt 10 Mbps) is exceeded, E2E
  delay increases due to the access network
• When access point starts discarding packets it
  causes increased E2E packet loss
• Internet delay is quite constant with all loads
  and packet loss negligible
• The access network is the bandwidth bottleneck
  and also the on with longest delay in stress
  situations

Figure 11. PanOulu delay performance with 16 Mbps
offered load.
Figure 12. Average delays and packet loss
12
Conclusions Future Work

• QoSMeT enables QoS measurement of IP networks
  from single applications point of view
• With multipoint measurement systems it is
  possible to locate performance bottlenecks in IP
  networks
• The multipoint measurement system combined
  terminal and intermediate point measurement
• The work is continued in Celtic Easy Wireless 2
  project
• Kick-off meeting 23-24 of September
• Light-weight measurement agents to the network
  for multipoint measurements
• The amount of measurement data that is collected
  and processed in real-time becomes problematic
• Application adaptation based on QoS measurements
  (e.g., SVC)

13
References

• ref1 J. Prokkola, M. Hanski, M. Jurvansuu, M.
  Immonen, Measuring WCDMA and HSDPA Delay
  Characteristics with QoSMeT, In proceedings of
  IEEE International Conference on Communications
  (ICC 2007).
• ref2 M. Jurvansuu, J. Prokkola, M. Hanski, P.
  Perälä, HSDPA Performance in Live Networks, In
  proceedings of IEEE International Conference on
  Communications (ICC 2007).
• ref3 P. Perälä, M. Jurvansuu, J. Prokkola,
  Combined Terminal and Network Measurement System
  for Bottleneck Localization, In proceedings of
  Tridentcom 2008.
• ref4 J. Prokkola, P. Perälä, M. Hanski and E.
  Piri, 3G/HSPA Performance in Live Networks from
  the End User Perspective submitted to IEEE
  ICC2009.

14
Convergent network and service infrastructure
that supports company RD with applied research
Benefits Practical, hands-on work with new
technologies and prototypes Innovation for new
product development New possibilities for
testing and prototyping on a live, "clean" and
highly configurable network Co-operation
Confidential company funded projects Publicly
funded research projects (EU/IST, Tekes, ITEA)
Licensing of VTT software Company products as
part of the CNL infrastructure Offerings
Research and consulting Testing and
prototyping Project preparations Networking