ops – an opportunistic networking protocol simulator for ...€¦ · networks based on mixim,...
Post on 05-Oct-2020
2 Views
Preview:
TRANSCRIPT
OPS – An Opportunistic Networking Protocol Simulator for OMNeT++
OMNeT++ Community Summit 2017 University of Bremen, Bremen, Germany
September 07 – 08, 2017
Asanga Udugama, Anna Förster, Jens Dede, Vishnupriya Kuppusamy and Anas bin Muslim
University of Bremen, Germany
Contents Motivation
Opportunistic Networks
Opportunistic Networking Protocol Simulator (OPS)
Evaluations
Summary and Future Work
2
Motivation Internet of Things (IoT)
Over 50 billion devices by 2020 [1]
Architecture for communications in the IoT Opportunistic Networking
IoT Scenarios Social networking to emergencies Nature of applications – higher value of information in locality
Importance of information propagation Forwarding protocols – Epidemic Routing, ODD, etc.
Necessity for large-scale evaluations Require simulators – OMNeT++
4
Characteristics of OppNets Information dissemination
Interested parties wanting information Value of information higher around the source
Store-and-Forward architecture Communicate when there is an opportunity to communicate Delayed delivery of information
Use of peer-to-peer communication technologies E.g., Bluetooth, IEEE 802.15.4
Importance of the information propagation Capabilities of the forwarding protocol
6
OppNets Use-case Propagation of information about an event
Street performers Interested people gather (flash crowd)
7
City Center
Visitor Street performersBuilding
time
Direction of messagesIntensity of the performance
Objectives Pluggable protocol layer architecture
Node model can handle new protocol implementations Clear interface between layers
Large-scale simulations IoT-scale devices
Mobility Synthetic, trace-based and hybrid
9
Protocol Stack Node model – 4 layer protocol stack
Protocol layers Application layer – Data generators Forwarding layer – Data propagation mechanisms Link Adaptation layer – Conversions to different link technologies Link layer – Link technology coupled with mobility
10
Application Layer
Opportunistic Forwarding Layer
Link Adaptation Layer
Link Layer Mobility
Models Application layer
Promote – Generates random data as constant traffic, uniformly distributed traffic or exponentially distributed traffic Herald – Generates pre-determined set of data where nodes assigned “likeness” value to data
Opportunistic forwarding layer Caching data – Employs store-and-forward Neighborhood communications – Communications with the changing neighborhood Epidemic Routing – Nodes negotiate and exchange data [2] Organic Data Dissemination (ODD) – Dissemination of data based on popularity of data [3] Randomized Rumor Spreading (RRS) – Random dissemination of data
11
Models …contd Link adaptation layer
PassThru – Simple packet traversal
Link layer WirelessInterface – Simple wireless interface that models bandwidth, delays, wireless range (with UDG) and queuing
Interfaces Use of an extensible packet format
12
Node Model Implementation An example node model used in an experiment
Use of trace based mobility BonnMotion – Cartesian trace of an actual GPS trace – SFO Taxi trace [4]
13
Evaluation Metrics Focus of performance evaluations is slightly different compared to classical networks
Data related metrics Liked Data – Preferred data received Non-liked Data – Not preferred but still received Traffic Spread – How well is packet traffic spread in the network Data Delivery Ratio – Delivery ratio of destined data Delivery Time – Delivery time of destined data
Mobility related metrics Average Contact Time – Duration of a contact Number of Contacts – Number of times in contact
14
Evaluation Scenario OPS is being used extensively in our research
Results of some evaluations Used in an IEEE Survey on OppNets [5]
General scenario details Nodes – 50-node network Mobility – SFO Taxi Trace [4] Data generation – 2 hour interval Run for 24 days
16
Influence of Traffic Models & Caching Scenario specific parameters
Different traffic generation models and different cache sizes Evaluation of data delivery times
Analysis Traffic generation model has no influence But, caching policy influences delay
17
0 h 6 h 12 h 18 h 24 h 30 h 36 h 42 h 48 h 54 h 60 hDelay
0.0%5.0%
50.0%
95.0%100.0%
CD
F(d
elay
)
Constant Tra�c, 100KB Cache Sizes
Constant Tra�c, 10KB Cache Sizes
Constant Tra�c, Infinite Cache Sizes
Constant Tra�c, 20KB Cache Sizes
Constant Tra�c, 50KB Cache Sizes
Poisson Tra�c, 100KB Cache Sizes
Poisson Tra�c, 10KB Cache Sizes
Poisson Tra�c, Infinite Cache Sizes
Poisson Tra�c, 20KB Cache Sizes
Poisson Tra�c, 50KB Cache Sizes
Uniform Tra�c, 100KB Cache Sizes
Uniform Tra�c, 10KB Cache Sizes
Uniform Tra�c, Infinite Cache Sizes
Uniform Tra�c, 20KB Cache Sizes
Uniform Tra�c, 50KB Cache Sizes
Performance of Mobility Models Scenario specific parameters
3 different mobility models (synthetic, trace-based and hybrid) Models parameterized as closely as possible to trace-based model
Analysis Trace-based takes the longest time (but realistic) Closest performance is given by the hybrid model (SWIM)
18
Fig. 4. CDF of the Delivery Delay of Data to Recipients using DifferentTraffic Generation Models and Cache Sizes
mobility of the nodes is performed using the BonnMotionmobility model of INET that is configured to use a 50-nodeextract of the San Francisco taxi cab trace (SFO trace) [11].The simulation time was 30 days and the applications deployedin each of these nodes generate data with a mean interval of2 hours.
Table I shows a comparison of the performance of thenodes in an OppNets network when using different mobilitymodels, viz., a simple random model, a trace based modeland a hybrid model implemented through Random Waypoint(RWP), Bonn Motion [12] and Small Worlds in Motion(SWIM) [13] [14], respectively. The Bonn Motion model usesthe SFO trace. As before, the network consist of 50 nodes anda simulation duration of 30 days.
Model RWP SWIM Bonn MotionSimulation Time 4 min 59 min 109 min
Memory used 74 MB 86 MB 127 MBAverage Delivery Rate 3 % 96% 92 %Average Delivery Delay 20.6 h 16.25 h 13.16 h
Total Number of Contacts 190 46,752 155,757Average Contact Duration 117.14 sec 150.12 sec 584.39 sec
Table I. Performance results of different mobility models consisting of asimple random model (RWP), a trace based model (Bonn Motion) and a
hybrid model (SWIM), with OPS
The applications deployed on the nodes generate data every2 hours and the data carry a specific destination to whichthe data must be delivered. Further, the RWP and the SWIMmodels are configured as closely as possible to represent thebehaviour of the SFO trace. The results show that mobility withreal traces require a higer level of resources (simulation clocktime and memory usage) compared to other mobility models.When considering the performance of the nodes, it is seen thatthe SWIM model provides the most closest results (AverageDelivery Rate, Average Delivery Delay, etc.) to the trace basedmodel. The trace based model is considered the most realisticdue to the traces being collected from real mobility patterns.
The survey [10] referenced before included the perfor-mance results of other widely used OppNets simulators [15],[16] in addition to OPS. The results show that OPS provides
a comparatively close performance to these simulators whenconsidering the metrics listed in Section IV.
VI. CONCLUSION
The Opportunistic Protocol Simulator (OPS) is an OM-NeT++ based modular simulator to evaluate the performanceof OppNets. The models in OPS implement the functionalityof the different layers of the protocol stack of an OppNetsbased node. The work presented here provides the details ofthe different models available in OPS.
OPS is used by us to simulate OppNets based scenarios[10] to evaluate the performance of dissemination protocols,traffic generation models, etc., to improve the performance ofOppNets in the IoT. OPS is released at Github under a GPLLicense (https://github.com/ComNets-Bremen/OPS).
OPS is being extended with new functionality on a regularbasis. Currently, there are a number of on-going projectsfocusing on areas related to forwarding protocols, applications,mobility models and user behaviour models. A long-term goalis to integrate OPS with the link technologies available withthe INET framework.
REFERENCES
[1] D. Evans, Cisco, The Internet of Things: How the Next Evolution of theInternet Is Changing Everything, April 2011
[2] A. Forster et al, A Novel Data Dissemination Model for Organic DataFlows, MONAMI 2015, September 1618, 2015, Santander, Spain
[3] L. Pelusi and A. Passarella and Marco Conti, Opportunistic Network-ing:Data Forwarding in Disconnected Mobile Ad Hoc Networks, IEEECommunications Magazine, November 2006
[4] A. Vahdat and D. Becker, Epidemic Routing for Partially-Connected AdHoc Networks, Technical Report, June 2000
[5] O. R. Helgason and K. V. Jonsson, Opportunistic Networking in OM-NeT++, SIMUTools 2008, March 03 - 07, 2008, Marseille, France
[6] O. R. Helgason and S. T. Kouyoumdjieva, Enabling Multiple Control-lable Radios in OMNeT++ Nodes, SIMUTools 2011, March 21 - 25,2011, Barcelona, Spain
[7] S. T. Kouyoumdjieva et al, Caching Strategies in Opportunistic Networks,WoWMoM 2012, June 25 - 28, 2012, San Francisco, U.S.A.
[8] R. Zhang et al, OppSim: A Simulation Framework for OpportunisticNetworks Based on MiXiM, CAMAD 2014, December 1 - 3, 2014,Athens, Greece
[9] Z. Zhao et al, Performance Evaluation of Opportunistic Routing Pro-tocols: A Framework-based Approach using OMNeT++, LANC 2012,October 4 - 5, 2012, Medellen, Colombia
[10] J. Dede et al, Simulating Opportunistic Networks: Survey and FutureDirections, IEEE Communications Surveys and Tutorials, Accepted forpublication in 2017
[11] Michal Piorkowski at al, CRAWDAD dataset epfl/mobility (v. 20090224),downloaded from http://crawdad.org/epfl/mobility/20090224, https://doi.org/10.15783/C7J010, February 2009
[12] N. Aschenbruck et al, BonnMotion - A Mobility Scenario Generationand Analysis Tool, SIMUTools 2010, March 15 19, 2010, Torremolinos,Malaga, Spain
[13] A. Mei and J. Stefa, SWIM: A Simple Model to Generate Small MobileWorlds, IEEE INFOCOM 2009, April 19 - 25, 2009, Rio de Janeiro,Brazil
[14] A. Udugama et al, Implementation of the SWIM Mobility Model inOMNeT++, OMNeT Community Summit 2016, September 15 - 16,2016, Brno, Czech Republic
[15] A. Keranen et al, The ONE Simulator for DTN Protocol Evaluation,SIMUTools 2009, March 2 - 6, 2009, Rome, Italy
[16] N. Papanikos et al, Adyton: A network simulator for opportunisticnetworks, [Online]. Available: https://github.com/npapanik/Adyton, 2015
Verification of the Models Survey compared OPS with 3 other OppNets implementations
ONE [6], Adyton [7] and ns-3
Analysis OPS provides a comparatively close performance (in metrics listed above)
19
Summary
21
OPS – OMNeT++ based modular simulator to evaluate the performance of OppNets
Node model architecture with pluggable protocol layers
OppNets focused evaluation metrics
Available at Github https://github.com/ComNets-Bremen/OPS
Future Work
22
Constant improvements, additions to OPS
Current projects Forwarding protocols (e.g. Spray and Wait) Applications User behavior models Mobility models
References
24
[1] D. Evans, Cisco, The Internet of Things: How the Next Evolution of the Internet Is Changing Everything, April 2011 [2] A. Vahdat and D. Becker, Epidemic Routing for Partially-Connected Ad Hoc Networks, Technical Report, June 2000 [3] A. Förster et al, A Novel Data Dissemination Model for Organic Data Flows, MONAMI 2015, September 2015, Santander Spain [4] Michal Piorkowski at al, CRAWDAD dataset epfl/mobility (v. 20090224), downloaded from http://crawdad.org/epfl/mobility/20090224, https://doi. org/10.15783/C7J010, February 2009 [5] J. Dede et al, Simulating Opportunistic Networks: Survey and Future Directions, IEEE Communications Surveys and Tutorials, Accepted for publication in 2017 [6] A. Keránen et al, The ONE Simulator for DTN Protocol Evaluation, SIMUTools 2009, March 2 - 6, 2009, Rome, Italy [7] N. Papanikos et al, Adyton: A network simulator for opportunistic networks, [Online]. Available: https://github.com/npapanik/Adyton, 2015
top related