Modeling traces is not so interesting now.

More exciting- spatio-temporal modeling of the Internet.

Would like to do topology modeling to infer: the paths of autonomous systems (AS) - AS paths; the traffic matrix at AS level not ISP level (e.g. use Akamai data).

Al Hero

Abilene Netflow data (slide) - non-linear relationship between quantities of interest.

Want to study :

• theory and practice of detecting rare events.

• theory of distributed collection/inference/control

• factorization/separation principles

• estimation/detection/communication/control

• 'local sufficiency' vs. 'global sufficiency'

• model reduction, parameterisation, compression, routing, decision information hiearchies

• large N,T,p regimes vs small N,T,p regimes

• proprietary and privacy concerns

• spatial/temporal independence and latency

Mark Crovella

Basic Science

• How many "computers" are connected to the Internet? IP addresses? routers? (Don't know but we should.)

• How many objects are accessible via HTTP? P2P? (We need to know this)

• How are these quantities changing over time and location? (We need to know this)

• Where are these computers and files?

Network Performance

• Real-time "hotspot" characterization: where, when and for how long do loss and congestion events occur in the wide area network?

• Network simulation: How can we use simulation to aid design? Also need to studytopology generation and source-modeled traffic generation.

Security

• How resistant is the Internet and its services to attack?

• What are the weaknesses?

• How strong are the adversaries? (search Google on "Now retired from spam fighting")

• How can we enforce protection of Internet resources? We need tools for tracing and identifying attacks.

Jean Meloche

Round trip times are easy; one way times are hard (need synchonization).

There are ways to partition time and estimate the clock relationships.

There are many problems that require knowing the topology- we need a methodology that takes data from a variety of sources and creates the topology.

Jianyan Sun

How do we root out emails sent by a hacker that uses my email address or a foreign language.

A possible solution is the use more clever filters that search the body of a messages and learn.

Mark Coates

Network Tracking

• How do you convey the "state" of the network?

• How do you define a "state"?

• Also need visualization tools.

Distributed Measurement

• Want to collect data at multiple sites, but where?

• How to deal with an enormous volume of data, high dimension data?

• How to deal with substantial spatial separation?

• Distributed Estimation & Detection

• How to encode data, OR how to generate summary statistics that permits data fusion? Want to map from high dimension to low dimension.

Distributed Learning

• How to identify behavioural changes through on-line training?

• How to determine when state changes occur?

Andre Broido

Internet Mapping

• Want to create a Point of Presence (PoP) level map of the Internet (the finest granularity we can hope to get right) .

• This is actually doable! (<6000 PoPs)

Traffic Measurements:

• How to make the most of high-precision backbone traces?

• Measuring trends- effects of IP applications, wireless, VoIP, streaming, games

Spectroscopy

• Interested in source identification for spoofed source addresses using delay quantization

Don Towsley

Interested in multi-level interactions from overlay networks like Akamai, Kazaa, gnutella

• Underlay control/overlay control: routing issues; what are the interactions of control loops?

• Underlay/overlay network measurements: can we understand the interactions?

• How do we make intelligent measurements to do intelligent control?