From chemical signals to event dissemination in a mobile system

1
From chemical signals to event dissemination in a
mobile system EPFL Distributed Programming
Laboratory Sébastien Baehni, Chirdeep S. Chhabra,
Rachid Guerraoui
2
Motivation
Devise an algorithm that implement the
topic-based publish/subscribe abstraction in a
mobile ad-hoc network environment
T2
T2
Topic hierarchy
T0
T1
T1
T2
T0
T0
3
Contribution

• The CSC2 (Chemical Signal Communication among
  Cells) Algorithm
• Uses the mobility of the processes and the
  validity periods of the events to enhance
  dissemination reliability
• Does not rely on any multicast or routing
  algorithm (one-hop communication)
• Saves memory by collecting old and noise events

NB Mobility Friendly Publish/Subscribe, S.
Baehni, C. S. Chhabra and R. Guerraoui, EPFL-LPD,
Technical Report No 200488.
4
Analogy

• Cells communicate by exchanging chemical signals
• The cells only react if they are sensible to
  these signals
• The signals have a limited duration and vanish
  after some time
• The signals reach only small geographical zones
  however, the cells move and hence propagate the
  information

Cell 1
Cell2
5
Roadmap

• Background
• CSC2
• Overview
• Evaluation

6
Background

• Dissemination algorithms
• Reliable multicast algorithms
• Make assumptions on the stabilization of the
  network
• Use cluster-heads
• Switch to flooding
• Counter/Distance/Location/Cluster-based
• Either use a GPS or are outperformed
• Neighboring schemes
• Dominant pruning (two hops neighbors)
• The publisher chooses the forwarding nodes
• Uses a greedy algorithm
• Self pruning (one hop neighbor)
• Each node takes the decision to rebroadcast or
  not
• Used in our algorithm (with modifications)

7
Background

• MANET Publish/Subscribe algorithms
• Use brokers (Siena, Jedi, )
• Use multicast trees
• Disseminate events according to a geographical
  location (Steam)
• Make network stabilization assumptions

8
Roadmap

• Background
• CSC2
• Intuition
• Evaluation

9
Overview

• Three phases
• Neighborhood Detection
• Event Dissemination
• Garbage Collection

Event Dissemination (I)
Neighborhood Detection
Interests
Events
Neighbors
Interests
Events
Neighbors
T1
e1
p2
T1
e1
p2
Interests
Events
Neighbors
Interests
Events
Neighbors
T1
---
p1
T1
---
p1
lt1gt
ltp1,T1gt
p1
p1
p2
p2
ltp2,T1gt
lt--gt
10
Overview
Garbage Collection
Event Dissemination (II)
Interests
Events
Neighbors
Interests
Events
Neighbors
T1
?
?
T1
e1
p2
Interests
Events
Neighbors
Interests
Events
Neighbors
T1
?
?
T1
e1
p1
lte1, p2gt
p1
p2
p2
p1
11
Neighborhood Detection

• Each process periodically broadcasts heartbeat
  messages (process identifier and interests)
• Once a process pi is detected by a process pk, pk
  puts pi in its neighborhood table if they share
  common interests
• If pi and pk do not share any common interest,
  they simply ignore each other

12
Event Dissemination

• Two new neighbors exchange their events
  identifiers according to their common interests
• Upon the reception of the events identifiers
• Each process checks if its neighbor misses an
  event
• The process sends the events (and the list of its
  neighbors) to its neighbors after a back-off
• The back-off is computed according to the number
  of events to send (the more the events to send,
  the smaller the back-off)
• Upon reception of an interested event by pi
• pi stores the event
• pi checks if it has to propagate the new event
  (self pruning)

13
Garbage Collection

• Neighborhood collection
• Each process periodically collects its
  neighborhood information
• Event collection
• The events are collected according to their
  validity and the number of times they have been
  propagated

14
Roadmap

• Background
• CSC2
• Intuition
• Evaluation

15
Evaluation

• Environment
• Qualnet simulator (initialized to the default
  802.11b values)
• Two models
• Random Waypoint Model (150 processes, 25km2)
• City Section Model (EPFL, 15 processes,
  1200x900m2, radio range of 44m)
• Speed of the processes and event validity vary
  for each experiment

16
Random Waypoint

• Reliability according to the speed of the
  processes, the validity of the event and the
  interests of the processes (20 versus 80)

17
Random Waypoint

• Reliability according to the validity of the
  event and the interests of the processes, in an
  heterogeneous environment

18
City Section

• Reliability according to the processes interests
• Difference of reliability between the processes
• Reliability according to the event validity

19
Summary

• CSC2 is used to disseminate events in a mobile
  ad-hoc environment
• Performance
• 90s validity is sufficient to ensure a 95
  reliability in the random waypoint model, where
  120 processes move at 30mps, in a 25km2
  environment
• The validity and the paths of the processes are
  extremely important in the City Section Model to
  let them go through hot points