The Gossiping Case Study

1
The Gossiping Case Study

• Boudewijn HaverkortMatthias KuntzLucia
  ClothNovember 2007Lorentz Centre, Leiden

2
Gossiping Program (Tuesday)

• 9.00 Keynote by Maarten van Steen, Uniform
  Randomness versus Emergent Behavior
• 10.30 Coffee break
• 11.00 Solution approaches (10 minutes each)-
  Rena Bahkshi, Free University Adam- Kai Lampka,
  ETH Zürich- Lucia Cloth, U Twente- Gianfranco
  Ciardo, UC Riverside- Jaco van de Pol, U
  Twente- Frits Vaandrager, Radboud U- Pepijn
  Crouzen, Saarland University
• 12.30 Lunch
• 14.00 Working session (14.00-16.00)
• 16.00 Group reports
• 17.00 Holger Hermanns, tQfQQ

3
Many Non-Functional Questions!

• Connectivity
• Node degree
• Clustering coefficient
• Path length
• Message complexity
• Information spread
• How do these properties change in the face of an
  accidental/massive node or link failure?
• How do these properties change in the face of
  node heterogenity?

4
MvS Experimental results

• Can we contribute to this, using model-based
  analyses?
• Can we avoid expensive/lengthy emulations or
  simulations?
• Can we just throw in our standard model
  checkers?
• Should we go to other methods, like from
  statistics or physics?

5
What did we discuss find ?

• Rehearsal of MDP/abstraction
• Rehearsal of mean field analysis (MFA)
• New ideas
• per node dynamics needs to be MDP(multiple
  communications per cycle)
• abstraction of MDP-based per-node model, such
  that it fits MFA
• Investigate which measures of interest cn be cast
  in the MFA approach

6
What did we discuss find ?

• MFA dynamical system (discrete-time)
• other models for dynamical systems (like ODEs,
  PDEs), like recently done for PEPA
• bottom line-- get away from enumerating
  things-- go to fluid-flow-like models-- will
  this save us from the state-space- explosion
  problem?-- can we solve the resulting eqns?