1. Network Science in NDSSL at Virginia Tech
2009 HPC User Forum
21 April 2009
Roanoke, Virginia

2. Two Examples
Wireless epidemics
Epidemic simulations

3. What people do to study this?
Wireless Epidemics
20.7 million smart devices were shipped in North America in 2007 and more expected in years to come
Expected growth of 150% in smart phone market
Ubiquity of smart digital devices lead to amplified opportunities for malware attacks
Open APIs result in buggy third party applications for smart devices
What people do to study this?
Use mathematical modeling to predict the impact
Useful and scalable, but not accurate
Simulation studies of the wireless networks with worm implemented
Accurate, not scalable (500 nodes -> 2 days)
We build models for studying them through a parallel discrete event simulator
Very little accuracy lost, but provides scalability ( > 1 hour)

4. EpiNet Simulation Framework
Activity-based mobility models device mobility
Sub-location modeling constructs wireless networks at activity locations
Malware manifestation models worm and protocol characteristics
What do we study with this?
Understand the spread
Detection strategies
Response schemes for handling outbreaks
Evaluate strategies
Conclusions
Network structure significantly alter the spread
More reason to consider realistic proximity networks
Early/late interventions are not that different
Dynamic interventions are required to be studied
Graphic of comparison of early versus late interventions with degree rank

5. Avian Influenza: The Threat
9 regional outbreaks in the last 11 years
Southeast Asia from
Economic loss of $10 billion US
62 deaths
Global concern: influenza pandemic
Possible fatalities: ~150 million deaths

6. Large Transmission System
poultry farms in the US
Farm sizes range from 1 to 5,000,000 birds
Avian influenza: avian-to-avian transmission among farms
One initially infected farm (node): #17196
~2500 miles

7. Simulated Virus Diffusion
time = t0
largest network size
nodes
414 million edges
initially infected node
(red nodes have contracted virus)

8. Vulnerability Maps for 3 Networks
Network M
Network L
phase transition
< 2000 infections
infections
> infections
community ≡ nodes within a geographic zone possessing high vulnerability

9. What Gives Rise To a Large Attack Size?
Red lines are boundaries between northern and southern communities
Boundary region is an area of low vulnerability
The virus must cross this divide for large attack size
The aforementioned divide is the dominant one, but there are others
Consequently, we have “speed up” and “slow down” of the virus spread
network M, only those simulations resulting in large attack size