Transcript Document
Fundamentals of Internet Measurement:
a Tutorial
Nevil Brownlee, Chris Lossley, “Fundamentals of Internet
Measurement: a Tutorial,” CMG journal of Computer Resource
Management, Issue 102, Spring 2001
정하경
2006.12.04
Presenter: Jung, Hakyung
Outline
Motivation
Measurement topics
H/W versus S/W approaches
Active versus Passive measurements
Examples
Main observations in the past research
Conclusion
2/15
Motivation
Network troubleshooting
Protocol debugging
Workload characterization
Test out “new, improved” version of network applications
and protocols
design of better protocols and network for supporting the
application
Performance evaluation and improvement
Identifying performance bottleneck
•
New versions of the protocols can provide better performance
3/15
Measurement Topics
Hardware approaches
Special-purpose tools designed to collect and analyze
network data
•
•
Often expensive
Better functionality and performance
Software approaches
Modify the kernel for the packet-capture capability
•
e.g. tcpdump
Access logs recorded by Web servers or proxies
4/15
Software approach example:
MIPv6 traffic measurement
CN
IPv6 Network
IPFIX
Flow Collector
IPv6 Router
IPFIX
flow data
HA
FA
FA
MIPv6 Access Router
with IPFIX
AP
MN
5/15
Measurement Topics
Passive measurements
Observe normal network traffic without creating additional
traffic
•
e.g. counting # of packets traveling through routers
Do not perturb the network
Rely on traffic flowing across the link being measured
Active measurements
Send test traffic needed to make the measurement
•
e.g. ping, traceroute utilities
Generate extra traffic onto a network, thereby affecting
measurement results
6/15
Passive techniques:
workload analysis (1/3)
A matrix of traffic from
source to destination
ASes
2 minute sample from FIXwest in April 1998
can be used to optimize
topology or to determine
the traffic balance among
ISPs
since AS is the unit of
routing relationship
7/15
Passive techniques:
workload analysis (2/3)
A matrix of traffic from
source to destination
countries
2 minute sample from FIXwest in April 1998
gives a perspective on
international commerce
8/15
Passive techniques:
workload analysis (3/3)
Cumulative distribution of
packet sizes, and of bytes
by the size of packets
carrying them
peaks at sizes of 44, 552,
576, and 1500 bytes
9/15
Active technique:
Internet mapping project
skitter, a tool for
dynamically discovering
and depicting global
Internet topology
uses the same procedures
with traceroute utility
gathers connectivity
information, RTT, path data
10/15
Active technique:
RTTs measurement
The distribution of RTTs
for 1600 probe packets
nearly symmetric &
relatively large distribution
The plot of delay values
measured along the path
congestion between
hop 11 and hop 12
11/15
Observations in the past
research (1/3)
Packet traffic is non-uniformly distributed
10% of the hosts account for 90% of the traffic
Client-Server paradigm
Network traffic exhibits “locality” properties
Packets are not independent; rather part of a higher-layer
logical flow of information
Temporal locality (time-based correlation)
Spatial locality (geography-based correlation)
12/15
Observations in the past
research (2/3)
The packet arrival process is not Poisson
Poisson arrival process
• Events occur independently at random times
• Inter-arrival times between events are exponentially
distributed and independent
• Memoryless property
Inter-arrival times between packets are not exponentially
distributed, nor are they independent
Packet arrival process is bursty
The session arrival process is Poisson
Users seen to operate independently at random
13/15
Observations in the past
research (3/3)
Traffic flows are bi-directional, but often
asymmetric
Download-intensive nature of the Web
Internet traffic continues to change
Traffic volume, traffic mix, protocols, applications, users
Data set collected from an operational network represents
bye one snapshot at one point in time in the evolution of the
Internet
14/15
Conclusion
Role of network traffic measurement
Design, testing, evaluation of Internet protocols &
applications
Internet traffic measurement Methodology
Summary of the main observations from the past
network measurement research
Internet traffic continues to change
15/15
Reference
N. Brownlee, C. Lossley, “Fundamentals of Internet
Measurement: a Tutorial,” CMG journal of Computer Resource
Management, Issue 102, Spring 2001
KC Claffy, “Internet measurement and data analysis:
passive and active measurement,” NAE workshop, 1999
C. Williamson, “Internet Traffic Measurement, “ IEEE Internet
Computing, Vol. 5, No. 6, 2001
http://www.cybergeography.org/atlas/topology.html
16/15