Design and Implementation of FlowLevel Simulator for Performance Evaluation of Large Scale Networks

1
Design and Implementation of Flow-Level Simulator
for Performance Evaluation ofLarge Scale Networks

• O Yusuke Sakumoto
• Ryouta Asai
• Hiroyuki Ohsaki
• Makoto Imase
• Graduate School of
• Information Science and
• Technology,
• Osaka University, Japan

2
Contents

• Background
• Packet-level simulator and flow-level simulator
• Research objective
• Flow-level simulator FSIM(Fluid based SIMulator)
• Experiment
• Simulation speed, accuracy, and memory
  consumption
• Conclusion and future work

3
Background

• Internet complexity
• The scale of the Internet has been expanding
  rapidly
• Difficult to understand its behavior
• ? Performance evaluation technique for
  a large-scale network has been demanded
• Performance evaluation techniques
• Mathematical analysis
• Simulation
• Experiment
• ? Simulation is the promising technique

4
Type of Network Simulators

• Packet-level simulator
• Mimic behavior of everypacket
• Have been widely used
• Flow-level simulator
• Mimic behavior of every flow
• Still under research

5
Packet-Level Simulator

• Advantage
• High accuracy comparedwith flow-level simulator
• Can measure packet-levelperformance metrics
• Disadvantage
• Inability to simulatelarge-scale networks

6
Flow-Level Simulator

• Advantage
• Fast simulation execution compared with
  packet-level simulator
• May simulate a large-scale/high-speed network
• Disadvantage
• Inaccuracy compared with packet-level simulator

a number of packets
a single flow
modeled
7
Related Work

• Propose an approach for large-scale network
  simulation4
• Numerical computational algorithm is quite simple
  
• Network states are updated at every fixed step
  time
• Network states are updated even when network
  states are unchanged
• ? Causing slowdown of flow-level
  simulation
• A TCP/RED fluid-flow model is utilized
• Do not model the TCP timeout mechanism
• ? Accuracy is not satisfactory

8
Research Objective

• Design and implement a flow-level simulator FSIM
• High accuracy and fast simulation execution
• Utilize detailed fluid-flow models13
• Adopt an adaptive numerical computation algorithm
• Compatibility
• Can input/output files compatible with ns-2
• Verify the effectiveness of FSIM
• FSIM outperforms conventional simulators

9
Adaptive Numerical Computation Algorithm

• FSIM uses adaptive stepsize control for the
  Runge-Kutta method15
• Adjusts the stepsize according to change in ODEs
• ? Computational complexity can be significantly
  reduced while maintaining the accuracy

window size
time
stepsize
10
Application of NumericalComputation Algorithm

• The TCP fluid-flow model requires past network
  states
• Past network states are required for
  calculatingthe next network state
• ? Save calculated network states in the
  memory
• Past network states might not have been saved
• FSIM uses the adaptive stepsize control
• ? Past network state is approximated
  by an interpolation of nearby network states

time
now
11
Compatibility with an Existing Performance
Evaluation Tool

• FSIM features
• FSIM can input/output files compatible with ns-2
• Input simple scenario files compatible with ns-2
• Output ns-2 trace files
• Trace-all/namtrace-all command
• FSIM advantages
• Can easily develop a simulation scenario file
• Can switch ns-2 and FSIM depending on network
  size
• e.g., FSIM ? Evaluation of large-scale network
  Ns-2 ? Evaluation of small-scale network

12
Experiment

• Evaluate the effectiveness of FSIM
• Simulation speed
• Accuracy
• Memory consumption
• Compare performance of three simulators
• FSIM
• ns-2
• FFM17 (a flow-level simulator)

13
Network Topology Used in Simulations
14
Experiment ResultSimulation Speed(1/2)
Increase
Hardly increase
15
Experiment ResultSimulation Speed(2/2)
65 of FFM
Execution time of FSIM is shorter than that of
FFM
16
Experiment Result Accuracy(1/2)
The simulation result of FSIM almost agrees with
that of ns-2
17
Experiment Result Accuracy(2/2)
FSIM can simulate faster than FFMwhile improving
the accuracy
18
Experiment ResultMemory Consumption
much less consumption
80 less consumption
FSIM can simulate much larger network than FFM
19
Conclusion

• Design and implement a flow-level simulator FSIM
• Verify the effectiveness of FSIM
• FSIM outperforms FFM
• The simulation execution time of FSIM is
  approximately 65 of that of FFM
• FSIM is high accuracy than FFM
• FSIM is 80 less memory consumption than FFM
• Our FSIM implementation is available at
• 
  http//www.ispl.jp/fsim/

20
Future Work

• Verify the effectiveness of FSIM
• Simulation of a large-scale network
• Support various types of networking protocols
• e.g., UDP, DCCP, HighSpeed TCP, and XCP
• Further improve the numerical computation
  algorithm
• Our FSIM implementation is available at
• 
  http//www.ispl.jp/fsim/