BlueTooth ScatterNet Performance Simulator BSPSFebruary 2002

1
BlueTooth ScatterNet Performance Simulator

• Supervisor Gil Zussman
• By Liron Har-Shai Ronen Kofman

2
Agenda

• What has been done so far
• Simulator capabilities.
• ScatterNet Routing Algorithm
• Results and analysis
• Conclusions
• Whats next

3
What has been achieved so far

• The simulator is up and running and achieving
  excellent results (compatible to calculations) .
• We are developed an algorithm to optimize packet
  flow in Scatternet with current capabilities of
  BT unit. The algorithm shows good result compare
  to other algorithms.
• We currently have a tool that allows a high
  quality analysis of ScatterNet behavior.

4
Simulator capabilities

• Create any PicoNet formation.
• Inner piconet routing.
• Delay measurement capabilities from every to
  every point.
• Scatternet with one bridge.
• Control each parameter of the of the packet flow
• Source types
• Arrival rates
• Policies
• etc .

5
Simulator capabilities
Master GUI
settings
Features

Source control
Policy control
6
Approaches to ScatterNet Routing

• Active Mode the bridge stays active in both
  PicoNets
• Sniff Mode bridge leaves one PicoNet and can
  return in sniff window.
• Hold Mode the bridge leaves the PicoNet for a
  certain time the master saves its AM_ADDR.
• We decided to explore the Hold Mode which gives
  a large space to develop an efficient algorithm.

7
Load Adaptive Algorithm

• The bridge decides when to switch according to
• Q size
• Maximum TimeOut
• Idle
• Time Commitment
• When bridge decides on a switch it sends Time
  Commitment (TC) to the master. This is the
  minimum time the bridge will be away. After TC
  the master has to be able to serve him

8
Results
Idle Q size TimeOut
We can see the effect of the load in the system
on the algorithms parameters. The algorithm
adaptiveness is demonstrated here
9
Results (cont.)

• Average arrival rate vs. average delay with
  different Max PicoNet Time
• In every load, there is one Max PicoNet Time
  that minimize the average delay.

Slave 6
Slave 1 (BD_ADDR 28)
10
Results (cont.)
We can see here the average delay in ScatterNet
using our Load Adaptive Algorithm (logarithmic
scale)
11
Results (cont.)
We can see here the average delay in ScatterNet
using our Load Adaptive Algorithm vs. the average
delay in ScatterNet using the Sniff Mode
algorithm (logarithmic scale)

• There are 2 graphs
• Max PicoNet Time 1.5 seconds and Sniff Interval
  48 slots
• Max PicoNet Time 3 seconds and Sniff Interval
  96 slots

12
Whats next

• Continue the results Analysis and reach final
  conclusions.
• Develop an optimal behavior of Scatternet
  regarding master and bridge policies.
• Define restrictions to feasible work.
• Add features to the simulator to make it as close
  to reality as possible.
• Publish all the above.