Transcript labview_ni_networks

Internet Measurement and
Analysis
Vinay Ribeiro
Shriram Sarvotham
Rolf Riedi
Richard Baraniuk
Rice University
Internet
• Complex interaction of
– Network protocols (routing, congestion control)
– Applications (Web, email, streaming video/voice)
• To improve performance
– Need to understand Internet’s working
– Need measurements and analysis tools
Network Operations: Performance
Issues
•Avoid network
congestion
–Monitor traffic
loads
•Fault tolerance
–Reroute traffic if
link goes down
• Must collect traffic statistics at routers/links
• Key: measurement must not overload routers
Measurement Granularity
• Packet level: measure every packet
• Connection level: measure
aggregate information for each file
transfer
• Link level: measure only mean bit
rate on link
data
volume
and
detail
What are performance relevant statistics?
Alpha/beta model for network traffic
• Motivation: study burstiness in traffic
• Connection-level analysis of available traces
• Finding: typically one dominant connection during
burst
• Classify dominant connections as alpha, remaining
as beta
5
x 10
x 10
5
x 10
3
2.5
2.5
2.5
2
=
1.5
1
0.5
0
0
2
+
1.5
1
0.5
2000
4000
6000
time (1 unit=500ms)
Overall traffic
0
0
number of bytes
3
number of bytes
3
5
2
1.5
1
0.5
2000
4000
time (1 unit=500ms)
6000
Alpha traffic
= all alpha connections
0
0
2000
4000
time (1 unit=500ms)
Beta traffic
= residual
6000
Implications of Alpha/Beta
• Network control
– Active Queue Management (AQM) should
target alpha connections that cause bursts
• Simulation: For realistic scenarios must
simulate both alpha and beta components
• Queuing analysis and network design
Should consider both components
– Gaussian beta component
– spiky alpha component
Network Users: Performance Issues
• Users want to optimize performance
– Use maximum bandwidth, minimize packet
delay and loss
• Internal routers uncooperative (security,
technical issues)
• Use probe packet delays measured by edge
machines
Available Bandwidth
• Available bandwidth = unused bandwidth on path
• Probing scheme must be non-intrusive (introduce
light load on network)
Chirp Probing
• Chirp: exponential flight pattern of probes
• Non-intrusive and Efficient: wide range of
probing bit rates, few packets
Estimation Algorithm
Available bandwidth = probing rate at onset of
queuing delay increase (congestion)
Applications
• Network path selection: download data over
path with more bandwidth
• Congestion control: find optimal rate to
transmit data without congesting network
Conclusions
• Internet measurement and analysis tools
required to improve performance
• Measurement modules needed at routers as
well as desktops
URL: spin.rice.edu