3.3 Simulation Models
3.3.3 Simulation Frameworks
The investigation of our proposed service overlay for harsh environments and disaster relief requires a versatile simulation framework. It must be able to simulate mobility of rescue workers and the communication between them including the underlying ad-hoc mesh net-work. State of the art simulation frameworks are usually dedicated to simulate either mobil-ity or communication systems. Usually, a mobilmobil-ity simulation framework is used to generate movement traces for all participants of a given scenario. These traces are then used as in-put for a network simulation framework to simulate communication systems exposed to the simulated movement. Unfortunately, an integrated simulation framework that provides both comprehensive mobility models and comprehensive communication models is yet missing.
We, therefore, present state of the art mobility and network simulation frameworks. We assess these frameworks according to the following requirements:
R.F1: Mobility models for disaster relief: The simulation framework should provide disaster-related mobility models as identified in the previous section.
R.F2: Online interface to other simulators: To generate realistic simulation results, an inte-grated simulation is necessary. The basis for such simulations is an online interface to another simulator so that multiple simulators can be connected.
R.F3: IEEE 802.11 WLAN model: Before, we argued that the underlay network in harsh en-vironments is based on an IEEE 802.11 WiFi mesh network. To realistically simulate communication using such an underlay, a realistic WLAN model is required.
R.F4: Peer-to-peer overlays: We base our peer-to-peer service overlay on the peer-to-peer paradigm. A simulation framework providing ready-to-use peer-to-peer overlays and tools tailored to developing new peer-to-peer systems is required for our investiga-tions.
Mobility Simulation Frameworks
For the simulation of mobility, a plethora of mobility simulation frameworks exists. There-fore, we only concentrate on tools especially designed for generating movement traces for research on mobile ad-hoc networks. A typical tool with such features is MobiSim [132].
Using different mobility models, it generates movement traces in XML format that can be used as input for network simulators. The tool CanuMobiSim [178, 182] further allows to consider obstacles within simulations. The framework Important [22] provides differ-ent mobility models and metrics to investigate the performance of routing protocols for ad-hoc networks. Their evaluation results show that the performance of different rout-ing protocols strongly depends on the mobility model used to generate movement patterns.
However, none of these frameworks generates realistic movement patterns for disaster relief scenarios.
Urban and Vehicular Mobility Frameworks
Furthermore, there exist several simulation frameworks for the simulation of urban mo-bility [111] and vehicular networks [64, 82]. The goal of these frameworks is to reproduce typical vehicular traffic for the investigation of car-to-car networks. Although, these simu-lation frameworks and their models reproduce realistic vehicular movement, they are not really suitable for disaster simulation. This has two reasons. First, these frameworks lack mobility models to reproduce movement of rescue workers during their mission. And sec-ond, as introduced earlier, the movement of first responder vehicles is mostly limited to the oscillation between two locations.
BonnMotion Framework
Most of the mobility models we introduced in Section 3.3.1 are implemented in the simu-lation framework BonnMotion [15]. It is a rich Java-based software that generates synthetic movement traces to investigate mobility in different scenarios. Furthermore, it can be used as a tool for the investigation of mobile ad-hoc network characteristics. The generated movement traces can be exported to be used in several supported network simulators. Espe-cially the implementation of the zone-based mobility model for disaster recovery [16] makes BonnMotion a suitable candidate for our investigations. Unfortunately, BonnMotion lacks an online interface for network simulators. Hence, an integrated simulation is not possible with this framework.
First Response Communication Sandbox
Finally, the first response communication sandbox (FRCS) by Bradler et al. [36] is a simu-lation framework supporting basic disaster and mobility simusimu-lation. It was developed at our group at the Technische Universität Darmstadt. The sandbox provides tools for simulating and visualizing disaster relief scenarios using a tile-based world model. Furthermore, the sandbox provides a preliminary interface for network simulators. In principal, this frame-work can be used for an integrated simulation. However, this frameframe-work lacks state of the art mobility models as compared to the BonnMotion framework, for instance.
Network Simulation Frameworks
For the investigation of communication systems, typically network simulators are used. Until today, a huge variety of such simulation tools exists. The network simulators differ in the design of their simulation engine as well as in the supported network types. Some simulators use a time-step-based simulation engine, such as PlanetSim [74]. PlanetSim uses a central step-clock to simulate timing. Therefore, the developer must decide how much time passes during one step of the central clock. Choosing a large time period for these steps results in less accurate results but at the same time allows to simulate large scale networks more efficiently. Vice versa, small time periods result in more accurate simulation results but simulating large scale networks consumes more time to execute simulation runs.
Most state of the art simulators use a discrete-event-based simulation engine [28, 105, 183, 192, 205]. Here, the simulated network nodes create and enqueue events for specific points in time. These events are processed by a scheduler chronologically without the need for a central clock. The currently processed event determines the current simulation time.
Because the developer does not need to define time steps using such a simulation engine
3.3 Simulation Models 63
design, one error source is eliminated. Hence, we concentrate on simulation frameworks based on a discrete-event simulation engine.
Low-level Network Simulation Frameworks
One of the first and most prominent network simulators is the Network Simulator version 2 (ns-2) [122]. It has become virtually the standard for network simulation [196]. Unfor-tunately, ns-2 has some significant shortcomings in the modeling details of the IEEE 802.11 modules. The successor 3 [85] is widely used today and will most probably replace ns-2 in the near future. The main goal of the ns-3 project is the improvement of simulation performance compared to ns-2. Both simulators are not compatible with each other requir-ing the portrequir-ing of ns-2 models into the ns-3 framework. Unfortunately, both simulators are tailored to simulate lower network layers in great detail. Hence, simulating large scale peer-to-peer systems is not the main focus of these frameworks. Also, both frameworks provide only few peer-to-peer overlays and moderate support for implementing new peer-to-peer systems.
Peer-to-peer Overlay Simulation Frameworks
Simulators, such as PlanetSim [74], OMNeT++ [192] and PeerfactSim.KOM [183], pro-vide means to simulate large scale peer-to-peer overlay networks more efficiently [134].
Especially for the design and investigation of new peer-to-peer systems, such frameworks provide many state of the art peer-to-peer overlays, such as Chord [185], Kademlia [121], and Pastry [163]. These overlays can be used as baselines in evaluations. The efficiency, however, results from an abstraction of the lower network layers. As our proposed service overlay for harsh environments is based on a wireless ad-hoc network, a detailed and correct model of the IEEE 802.11 WLAN standard is required. The most promising candidate for our evaluations is the PeerfactSim.KOM framework. The developers have ported the realistic WLAN model based on the IEEE 802.11g standard specifications [90] from ns-3 into Peer-factSim.KOM to facilitate the simulation of mobile peer-to-peer and ad-hoc networks [184].
This enables us to investigate our proposed service overlay under realistic wireless network conditions.
Results
We present our results in Table 6. In general, simulation frameworks providing online in-terfaces for other simulators are not particularly common. It is not possible to simulate so-phisticated behavior-based mobility and communication models together in one integrated simulation framework, yet. The only framework providing a preliminary interface to other simulators is the first response communication sandbox by Bradler et al. [36]. However, the mobility models provided in that framework are rather simple and cannot keep up with the mobility models implemented in the BonnMotion [15] framework.
Network simulation frameworks, on the other hand, generally lack mobility models for disaster relief scenarios. They concentrate on the simulation of network-related issues and only provide very basic and often random-based mobility models. The only simula-tion frameworks providing a realistic IEEE 802.11 WLAN model are the ns-3 [85] and the PeerfactSim.KOM [183] simulators. The ns-3 simulator, however, lacks a rich portfolio of
Mobilitymodels fordisasterrelief Onlineinterfaceto othersimulators IEEE802.11 WLANmodel Peer-to-peer overlays
Requirements R.F1 R.F2 R.F3 R.F4
Mobility simulation frameworks
MobiSim [132] (Ø) − − −
CanuMobiSim [178, 182] (Ø) − − −
Important [22] (Ø) − − −
SUMO [111] − − − −
VanetMobiSim [64, 82] − − − −
BonnMotion [15] Ø − − −
FRCS [36] (Ø) Ø − −
Network simulation frameworks
ns-2 [122] − − (Ø) (Ø)
ns-3 [85] − − Ø (Ø)
PlanetSim [74] − − − Ø
OMNeT++ [192] − − (Ø) Ø
PeerfactSim.KOM [183] − − Ø Ø
Table 6: Evaluation of presented simulation frameworks
peer-to-peer overlays and comprehensive tools to implement peer-to-peer systems. The best fit for our investigations is, therefore, the PeerfactSim.KOM simulator.
3.4 Summary
In this chapter, we investigated related work for the three major fields of this thesis: peer-to-peer-based service provision, harsh environments (the communication infrastructure in particular), and simulation models. First, we evaluated communication infrastructures for harsh environments and corresponding routing protocols. Our investigation of communi-cation infrastructures revealed that a decentralized communicommuni-cation infrastructure based on the WiFi standard would be the best fit for disaster relief. It outperforms existing standards in terms of throughput, openness and failure resilience. Pre-deployed infrastructures are readily available in disaster situations and can be further extended with roll-out infrastruc-ture. Unfortunately, pre-deployed infrastructures are not widely available. Only few wireless mesh networking projects exist and only in big cities. A more omnipresent infrastructure is yet missing. In terms of routing protocols, we identified reactive protocols to provide higher performance for highly dynamic mobile ad-hoc networks. AODVv2 [43, 156] is the most advanced reactive routing protocol and the best candidate for harsh environments as of now.
3.4 Summary 65
Second, we evaluated systems for distributed service provision. We presented the code mobility concept and the mobile agent paradigm which constitute the basis for service mo-bility in our peer service overlay concept. We evaluated general concepts for peer-to-peer-based service provision as well as concepts from grid and cloud computing to be used in harsh environments. Although no concept was able to fulfill the requirements for service provision in harsh environments, the cloudlet concept [109, 169] is the most promising can-didate to be considered in our system design. Cloudlets are nearby resource-rich computing nodes (e.g., PCs connected to wireless routers in cafés or bars) that could be used for of-floading complex computing tasks from the mobile device. We further investigated service placement and service discovery. In terms of service placement, most approaches relocate the service instance to reduce network traffic. None of the investigated approaches consider quality of the provided service as metric to relocate the service instance. Therefore, we identified the best performing service placement approach for traffic reduction to be used as our evaluation baseline. This approach is called tree placement [142]. In terms of ser-vice discovery, most approaches were designed and evaluated to provide good performance under peer mobility. Service mobility was not considered in most cases. We, therefore, se-lected three discovery approaches to be evaluated for their performance in combination with service placement: Directory Backbones [165], Service* [135], and GSD [45]. Finally, we briefly investigated the remaining system components and identified suitable approaches to be used in a real world implementation of service overlays.
Third, we investigated simulation models and tools for harsh environments. We investi-gated mobility and communication models to be used as load factors for simulation-based evaluations. To reproduce realistic movement of all participants in harsh environments (including civilians), a more sophisticated mobility model is required. We found that a com-bination of zone-based [16], role-based [136], and gravity-based [136] mobility models would produce the most realistic movement patterns for simulations of harsh environments and of disaster relief scenarios in particular. Furthermore, we identified the conversation model for disaster areas [17] and a slightly modified version of the G.729 audio codec [95]
to realistically generate communication load for disaster recovery workers. Finally, we eval-uated mobility and network simulation frameworks for their suitability as evaluation tools for harsh environments. Unfortunately, an integrated framework that provides versatile mo-bility models for disaster relief and also allows the realistic simulation of WLAN networks is yet missing. However, we identified the BonnMotion [15] framework to provide suit-able disaster-related mobility models and the first response communication sandbox [36] to provide an online interface to connect a network simulation framework. For an integrated disaster simulation we further identified PeerfactSim.KOM [183] to provide the required WLAN model and tools for the development of peer-to-peer systems.
4 Communication Infrastructure
To provide versatile software services in harsh environments, a resilient communication sys-tem is required. In harsh environments, however, the communication infrastructure is often destroyed or overloaded (cf., Chapter 2). Furthermore, today’s standards exhibit single points of failures and can only deliver low data throughput (cf., Section 3.1). We, there-fore, propose a solution based on decentralized communication schemes on top of the IEEE 802.11 WLAN standard. The contributions in the present chapter are threefold:
1. We propose our concept of the CityMesh. It is a highly distributed emergency mesh network built from private and public wireless routers in inhabited areas that would be used for disaster relief communication.
2. We present our methodology to evaluate the feasibility of such CityMesh networks by collecting and analyzing data about wireless routers in cities. We further describe a method to estimate real router locations from measured data sets.
3. We evaluate the feasibility of CityMesh networks in several cities and show the severe improvement of network resilience if private routers are used in CityMesh networks.
The contributions in this chapter were published in the proceedings of the international Information Systems for Crisis Response Management (ISCRAM) conference in 2011 [152]
and in 2012 [150]. Furthermore, the contributions were published and in the International Journal of Mobile Network Design and Innovation (IJMNDI) in 2012 [151] and in the In-ternational Journal of Information Systems for Crisis Response Management (IJISCRAM) in 2012 [153]. In addition, our proposed approach attracted international media attention in the second half of 2012. Articles were published on several news websites, such as Ars Technica9, Deutsche Welle10, and Heise Online11.
The remainder of this chapter is organized as follows: In Section 4.1, we present our CityMesh concept in detail and how it could be implemented in the future. We present our methodology for evaluating the feasibility of CityMesh networks in Section 4.2. In Sec-tion 4.3, we evaluate the feasibility of CityMesh networks in five selected cities. Afterward, we discuss realization aspects in Section 4.4 and summarize the chapter in Section 4.5.