Vern Paxson and Sally Floyd, "Why We Don't Know How To Simulate The Internet", Proceedings of the 19

1

• Vern Paxson and Sally Floyd, "Why We Don't Know
  How To Simulate The Internet", Proceedings of the
  1997 Winter Simulation Conference, Dec1997
• Sally Floyd and Vern Paxson Difficulties in
  Simulating the Internet, 2001
• Realistic BGP Traffic for Test Labs. Olaf Maennel
  and Anja Feldmann, SIGCOMM 02
• Network Topology Generators Degree-Based vs.
  Structural. H. Tangmunarunkit, R. Govindan), S.
  Jamin, S. Shenker and W. Willinger, SIGCOMM02

2
Motivation

• Simulating how the global Internet behaves is
  important
• Validate new protocols before deploying them
• Immensely hard task
• Sheer diversity of the hosts
• Internet is big
• Jan 1996 - 9 million computers
• Jan 1997 - 16 million computers
• Dec 2000 - 100 million computers
• Internet changes in drastic ways
• Growth (80) can be traced back to 1984

3
Motivation (cont)

• Growth affects usage statistics with time
• ftp median size went from 4500 bytes to 2100
  bytes!!
• New protocols take over the traffic
• Web traffic was unknown before 1992, Napster,
  Gnutella, Mbone grew on short notice. Napster
  traffic disappeared overnight
• What are we trying to achieve with simulations
• Faithfully model the internet or capture its
  global scale behaviour?

4
Heterogeneity

• Topology and link properties
• What is the network topology - hierarchical or
  power law distributions
• Network is also evolving continuously
• Link properties also change (DSL, cable, and
  other technologies)
• Links can be asymmetric - especially with large
  topologies
• Protocol differences
• TCP implementations are evolving too (with their
  own set of bugs!!)
• Depends on the application goals for simulation

5
Traffic generation

• How do we generate representative traffic
• Trace-driven simulation is tempting
• Internet uses adaptive congestion control
• Traces depend on the time when it was collected
• Trace driven source-level simulation is promising
• Todays network is not tomorrows
• New pricing structure could make QoS take off
• Routing protocols may introduce QoS models
• Native multicast rather than tunnelled
• CDNs changed the web dynamics
• Next killer application can change everything

6
Coping stategies

• Search for invariants
• Carefully exploring the parameter space
• Choosing a random topology does not prove
  anything
• Need to analyze fro a wide range of viable
  parameters!!

7
Other papers

• BGP traffic generators
• Network topology generators