Juwon, park

1
End-to-end Performance Enhancement based on
Hybrid Network Monitoring for Continuous Media
Delivery

• Juwon, park
• 17, Feb. 2003
• jwpark_at_kjist.ac.kr
• Networked Media Lab.
• Dept. of Information Communication
• Kwang-Ju Institute of Science Technology
• Republic of Korea

2
Outline

• Introduction
• Related work
• Proposed scheme ( Hybrid monitoring )
• Experiment
• Conclusion

3
Introduction

• Traditional Internet Application
• Web, FTP, e-mail, telnet, etc.
• Emerging multicast Internet application
• On-line game, conferencing, Internet broadcasting
• VOD service, VoIP
• In order to guarantee QoS, Measurement is basic
  technique
• Measure not only network condition but also
  application state

4
Related work

• B.B. Lowekamp
• Combining active and passive network
  measurements to build scalable monitoring systems
  on the grid." Performance Evaluation Review,
  30(4) 19-26, 2003.
• By combining active and passive monitoring
  technique, he is able to reduce the need for
  invasive measurements of the network without
  sacrificing measurement accuracy on WAN
• Basic idea
• When a grid application is running, all
  measurements are made passively. However, when no
  application is running, or when the grid
  application is not sending sufficient traffic
  between the desired pairs of machines, we use
  active bandwidth probes

5
Related work

• B. Landfeldt
• The Case for a Hybrid Passive/Active Network
  Monitoring Scheme in the Wireless Internet" in
  Proceedings of IEEE international conference
  networks, 2002
• This paper propose hybrid monitoring scheme which
  combine active and passive network monitoring
  scheme in the Wireless Internet.
• Mechanism
• The mobile host transmits a request packet to the
  EXPAND (extended passive monitoring server)
  server.
• EXPAND server will respond with the requested
  information about the wired network segment.
• The request-response packets will be used by the
  requesting mobile host as active probes to
  determine the characteristics of the wireless
  access network
• gt By using both requested information and
  request-response packet, mobile host can estimate
  the end-to-end characteristics of the connection.

6
Related work

• M. Aida et.al
• "CoMPACT-Monitor Change-of-measure based
  passive/active monitoring weighted active
  sampling scheme to infer QoS," in Proceedings of
  Applications and the Internet (SAINT) Workshops,
  2002.
• This paper proposes a new performance measurement
  scheme which can infer performance and/or quality
  experienced by individual user, organization and
  application via a scalable and light-weight
  measurement way
• By using result of passive monitoring scheme, we
  can improve accuracy of active monitoring scheme
  (CoMPACT change of measure based no
  passive/active monitoring scheme)

7
Metrics

• Target metrics
• Availability The network availability
  characteristic is a measure of whether packet can
  reach destination or not.
• delay (RTT) RTT is measured by comparing send
  time and receive time.

Probe packet
LSR
Current Time
Probe packet
SR
SR or RR
Sender
Receiver
DLSR
Network RTT
Application RTT RTT Current time DLSR LSR
8
Metrics

• Jitter
• The delay variation
• Important to streaming application Prebuffer
  size
• Loss
• Repeated loss
• The statistical properties for determining
  application performance.

9
Proposed Scheme ( Hybrid monitoring )
Application layer
Problem point Decision Module
Encoding
Packet size
FEC
Application? Network?
ARQ
Rate control
Priority
RTP
Audio/Video Data
RTCP
Application Layer
Network Layer

RTCP Packet capturing
RTP
Probe Packet
Multicast network
Active Monitoring Client
??
Active monitoring module
RTP

• Jitter

• RTT

• Loss

RTCP

• RTT

RTP
Access Grid Node
10
Hybrid monitoring scheme

• Proposed monitoring scheme
• Use both active monitoring and passive monitoring
  scheme
• Network environment multicast network
• Active monitoring scheme
• Measure network condition by using multicast
  probe packet
• Metrics delay, loss, jitter
• Passive monitoring scheme
• Measure application condition by capturing RTCP
  packet
• RTCP provide feedback on the quality of the data
  distribution
• Can reduce system overload by capturing RTCP
  packet

11
Hybrid monitoring components

• Hybrid Monitoring Server
• Collect the both network and application metrics
  information
• Save data into file
• Show result to user by using Web
• If user want to see the past result, then show
  result to user by using JAVA applet

12
Hybrid Monitoring Client

• Active Monitoring Module (Network layer)
• Network layer metrics ( loss, jitter, RTT )
• Multicast probe packet

calculate delay, jitter
Classify sender
calculate loss
probe packet format
Sender info.
Sequence
Time stamp
Sender Addr.
Dest. Addr.(Multicast addr.)

• Passive Monitoring Module (Application layer)
• Application layer metrics ( loss, jitter, RTT)
• Capture RTCP packet instead of RTP

Flow chart of passive monitoring module
13
Previous VS. Current
Source
Previous version
Current version

• Classify the result based on source
• Detect the leave/join participants
• Discriminate result easily among participants

14
Experiment Environment

• Organize two network by using 2621_Router
• Multicast enable
• execute Access Grid program at NML and jwpark
• AG using RAT and VIC (multicast packet)
• Impose heavy load on network by using Iperf

15
Analysis
Passive Case
System A
System B
Heavy load
Network
d_system_a_make_packet d_network
d_system_b_receive
d_system_a_receive_packet d_network
d_system_b_send
Total Application RTT d_system_a_make_packet
d_system_a_receive_packet 2d_network
d_system_b_receive d_system_b_send
Active Case
Total Network RTT 2d_network
16
Analysis

• System overload case
• Network condition is normal (active result is
  similar to normal condition)
• d_network factor equal normal condition result
• The most affective factor is d_system_a_make_pack
  et and d_system_a_receive_packet

By comparing active result and passive one, it is
possible to find out problem point
17
Result Figure

• In system problem case
• Application RTT increased
• Network RTT is similar to normal case

18
Solution

• Proposed solution
• The problem factor
• Too mach system load
• Solution Reduce system load
• Do not apply the FEC scheme
• Do not apply the adaptive play-out scheme
• Do not apply redundancy control
• Reduce the packet size

19
Analysis
Passive Case
System A
System B
Heavy load
Network
d_system_a_make_packet d_network
d_system_b_receive
d_system_a_receive_packet d_network
d_system_b_send
Total Application RTT d_system_a_make_packet
d_system_a_receive_packet 2d_network
d_system_b_receive d_system_b_send
Active Case
Total Network RTT 2d_network
20
Analysis

• Network overload case
• The most affective factor is d_network
• The system_a_make packet, d_system_b_receive,
  d_system_b_make_packet, d_system_a_receive is
  very small gt ignore

By analyzing active result , it is possible to
find out problem point
21
Result Figure

• Network RTT increased
• Application RTT increased because d_network
  factor affect application RTT.

22
Solution

• Proposed solution
• The problem factor
• Too mach network load
• Solution By using application scheme
• Apply the interleaving scheme
• Apply the piggybacking scheme
• Apply the adaptive play-out scheme
• Apply FEC module
• No retransmission

23
Conclusion

• For guaranteeing QoS, need to measure exactly not
  only network condition but also application
• By comparing and contrasting the active and
  passive measurements
• the co-validity of the different measurements can
  be verified
• Get the much more detailed information
• Point out the problem layer ( network layer? Or
  application layer?)
• Future work
• Estimate available bandwidth by using packet pair
  mechanism
• Control the interval of probe packet based on
  passive result

24
Reference

• 1 B. Lowekamp, et al., "A hierarchy of network
  measurements for Grid applications and services
  (DRAFT),"  GGF NMWG internal Draft, 2002.
• 2 Access Grid, Available http//www.accessgrid.
  org/.
• 3 RAT, Available http//www-mice.cs.ucl.ac.uk/m
  ultimedia /software/rat.
• 4 V. Hardman, A. Sasse, M. Handley, and A.
  Watson, "Reliable audio for use over the
  Internet," in Proceedings of INET'95, June 1995.
• 5 I. Kouvelas, O. Hodson, V. Hardman, and J.
  Crowcroft, "Redundancy control in real-time
  Internet audio conferencing," in Proceedings of
  AVSPN 97, 1997.
• 6 C. Perkins, I. Kouvelas, O. Hodson, V.
  Hardman, M. Handley, J. Bolot, A. Vega-Garcia,
  and S. Fosse-Parisis, "RTP payload for redundant
  audio data," IETF       RFC 2198, September 1997.
  
• 7 Beacon, Available http//dast.nlanr.net/Proje
  cts/Beacon/.
• 8 H. Schulzrinne, S. Casner, R. Frederick, and
  V. Jacobson "RTP A transport protocol for
  real-time application," IETF RFC 1889, January
  1996.
• 9 S. McCanne and V. Jacobson, "vic A flexible
  framework for packet video," in Proceedings of
  ACM Multimedia '95. 1995.
• 10 M. Aida, K. Ishibashi, and T. Kanazawa,
  "CoMPACT-Monitor Change-of-measure based
  passive/active monitoring weighted active
  sampling scheme to infer QoS," in Proceedings of
  Applications and the Internet (SAINT) Workshops,
  2002.
• 11 Iperf, Available http//dast.nlanr.net/Proje
  cts/Iperf1.1.1/.