Designing a protocol for active network measurement

1
Designing a protocol for active network
measurement

• CS744 Future Internet Final Presentation
• 10 December, 2007

Changhyun Lee Advanced Networking Laboratory
KAIST
2
Outline

• 1. Introduction
• 2. Background
• 3. Motivation
• 4. Project Goal
• 5. Requirements
• 6. Solution Approach
• 7. Experiment
• 8. Future Work
• 9. Conclusion

3
1. Introduction

• Internet measurement is critical in assessing
  current architecture and designing for future
• Two types of measurement exist
• Active
• Passive
• Active measurement is more widely used

4
2.1. Performance Metrics

• Latency
• How long it takes to get response from a host
• Bandwidth
• How much data can be delivered per time unit
• Loss rate
• How much portion of packets are dropped at each
  router

5
2.2. Current Methods

• Latency measurement
• Uses TTL-limited ICMP, UDP packets
• Tools Ping, Traceroute Jacobson89
• Bandwidth measurement
• Saturates links or uses packet pair method
• Tools Pathchar Jacobson97, Spruce Strauss03
• Loss rate measurement
• Simple sending of dummy packets

6
2.3. Weaknesses of Current Methods

• Waste of measurement traffic
• Probe packets are needed as many as the number of
  hops
• Inaccuracy of result
• Different probe packets experience different
  network conditions
• Calculation is used rather than direct
  measurement
• Limited coverage
• Only source node initiates active measurement
• Vantage points are necessary

7
3. Motivation

• Active measurement was not considered when the
  current architecture was designed
• It limits the quality of measurement
• Many problems still exist despite much effort
• In future Internet, architectural support for
  active measurement will be necessary

8
4. Project Goal

• Design of a protocol for active network
  measurement, which can be a key component in
  future Internet architecture

9
5. Requirements

• Before designing the protocol, it is important to
  know what would be required in the future
• Weaknesses of current scheme must be overcome
• Main requirements
• Accuracy
• Wide coverage
• Robustness

10
5.1. Accuracy

• Active measurement result should represent the
  network condition that real traffic experiences
• Measurement should be done during one routing
• Hop-by-hop performance should be directly
  measured, not calculated

(O)
5
6
1
(X)
11
5.2. Wide Coverage

• Anyone should be able to measure the performance
  of arbitrary paths without need of vantage points
• Measurement should cover all the hosts in the
  Internet
• No more inferences will be needed
• Large scale performance study will be possible

12
5.3. Robustness

• No one should be allowed to harm routers and
  end-hosts by abusing the new protocol
• Time limits to prevent too much overhead must be
  required

13
6. Solution Approach

• Basic concepts
• User sends a measurement request packet to any
  host to start from
• Measurement source create a probe packet and send
  to the destination
• Probe packet triggers measurement at each router
  as routed to the destination
• Each router or host sends the result back to the
  user

14
6.1. Latency Measurement

• When a probe packet arrives at each host, the
  time that the packet was received is sent back to
  the source
• The source host gathers all the response and
  determine the latency

15
6.1. Latency Measurement (contd)
Probe Dst address Latency
12 ms
56 ms
23 ms
Hop1
Src
Hop2
Dst
Hop1 120523.412000
Hop2 120523.468000
Dst 120523.491000
16
6.2. Bandwidth Measurement

• Each host performs bandwidth measurement to the
  next-hop, when a probe packet arrives
• To measure one-hop bandwidth, traditional methods
  can be used
• Each host sends the result back to the source

17
6.2. Bandwidth Measurement (contd)
Probe Dst address Bandwidth
70Mbps
100Mbps
80Mbps
Hop1
Src
Hop2
Dst
Src-gtHop1 70Mbps
Hop1-gtHop2 100Mbps
Hop2-gtDst 80Mbps
18
6.3. Loss Rate Measurement

• Each host that receives a probe packet sends
  dummy packets to the next-hop and see how many
  are dropped
• Each host sends the result back to the source

19
6.3. Loss Rate Measurement (contd)
Probe Dst address Loss Rate
0.05
0.03
0.10
Hop1
Src
Hop2
Dst
Src-gtHop1 0.05
Hop1-gtHop2 0.03
Hop2-gtDst 0.10
20
7.1. Experimental Setting

• A virtual path on PlanetLab Peterson02
• Path of 12 nodes from UMass to UBC
• 10 nodes in the middle are regarded as routers
• As a first step, the implementation supports
  latency measurement

21
7.2. Experiments
request
22
7.3 Preliminary Result

• planetlab1.cs.umass.edu
• planetlab1.cs.cornell.edu
• planetlab1.eecs.umich.edu
• ricepl-2.cs.rice.edu
• planetlab1.csres.utexas.edu
• planetlab1.utep.edu
• planetlab-dsl-1.ucsd.edu
• planet1.cs.ucsb.edu
• planetlab2.ece.ucdavis.edu
• planetlab1.cs.uoregon.edu
• planetlab01.cs.washington.edu
• planetlab1.cs.ubc.ca

10.508ms
25.757ms
8.834ms
8.668ms
8.199ms
50.776ms
248.159ms
0.421ms
27.589ms
23.337ms
23.009ms
23
8. Future work

• To see how accurate result the new protocol can
  provide
• Comparing to the traditional methods
• To specify more detail on robustness

24
9. Conclusion

• Active measurement should be supported in the
  future Internet
• The design in this work is a first step to
  develop a protocol for active measurement
• Experiments show that the design is practical
• Both accuracy and coverage are improved

25
References

• Jacobson89 Van Jacobson, traceroute, February
  1989, ftp//ftp.ee.lbl.gov/traceroute.tar.gz.
• Jacobson97 Van Jacobson, Pathchar,
  ftp//ftp.ee.lbl.gov/pathchar/.
• Peterson02 Larry Peterson et al., A Blueprint
  for Introducing Disruptive Technology into the
  Internet, Proceedings of the First ACM Workshop
  on Hot Topics in Networking (HotNets), October
  2002.
• Strauss03 Jacob Strauss et al., A measurement
  study of available bandwidth estimation tools,
  Internet Measurement Conference, 2003
• Claffy05 KC Claffy et al., Community-Oriented
  Network Measurement Infrastructure (CONMI)
  Workshop Report, GENI Design Document 06-40,
  December 2005.
• Barford06 Paul Barford et al., GENI
  Instrumentation and Measurement Systems (GIMS)
  Specification, GENI Design Document 06-12,
  September 2006.
• Shalunov06 Stanislav Shalunov et al., One-way
  Active Measurement Protocol, RFC 4656, September
  2006.
• Madhyastha06 Harsha Madhyastha et al., iPlane
  An Information Plane for Distributed Services,
  Symposium on Operating System Design and
  Implementation, November 2006.
• Spyropoulos07 Thrasyvoulos Spyropoulos et al.,
  Future Internet Fundamentals and Measurement,
  ACM SIGCOMM Communication Review, volume 37,
  April 2007