Clustering of Data Mules for Faster and Reliable Data ... - satnac

Clustering of Data Mules for Faster and Reliable Data Delivery in Isolated Networks Xoluqobo Mkhwanazi , Hanh Le and Edwin Blake Department of Computer Science University of Cape Town, Private Bag X3, Rondebosch, 7700 Tel: +27 21 6502663, Fax: +27 21 6503551 Email: {xmkhwanazi, hanh, edwin}@cs.uct.ac.za Abstract- In this paper we propose a new algorithm, called Data Mule Intercommunication (DMI), that aims to increase message delivery between isolated networks. A data mule is a combination between an electronic device and a mobile entity. Mules are used to deliver messages between networks that were disconnected traditionally. This paper proposes to cluster data mules in order to find a more reliable and faster path between the sender and the receiver. Instead of data remaining on the same mule when travelling from one network to another, our algorithm allows mules to transfer data to nearby mules arriving to the destination network sooner. A variation of our algorithm is a combination of clustering between data mules and the optimal relay path (ORP) algorithm. The ORP calculates the shortest path from one data mule to another. Index Terms—data mules, networks, isolated networks.

clustering,

ad

hoc

I. INTRODUCTION Simple ideas make a world of a difference. One of these ideas was the data mule. It’s a combination of any data carrier and mobile entity. This means a data mule could be a person with a USB stick or a laptop powered by a bus. The data mule has the purpose of picking up data from one access point, buffering it and dropping it off at another access point [1]. We can use mules to connect isolated networks. Isolated Networks are networks that are out of range of each other and this may be due to lack of infrastructure or terrain. We found that by using mules to connect isolated networks there was long latency experience. The main reason for this latency is because the user of the mule has no control on the direction or speed of the mule [1,2]. This means when you send a message using a mule, the time taken to send from one point to another is dependent on how it will take the mule to arrive at the destination. In our research we aim to reduce the latency experienced when sending data using mules to connect isolated networks. In order to this we are developing an algorithm that will use clustering of multiple mules for faster communication and find the shortest path from the sender network to the destination network faster. Furthermore the algorithm it aimed to be applied in a rural area. We will briefly discuss related work and the algorithm development.

II. RELATED WORK By using mules to connect isolated networks a delay tolerated network (DTN) is formed. DTNs are disconnected networks, where transmission of information across the network is subject to long latency sometimes measured in days or hours [3,4]. A good example of a DTN that uses a mule for connectivity is the DakNet project in India. A. DakNet The DakNet project [5], uses mules to connect surrounding isolated village networks. The mule is a bus and a laptop charged by the bus. It travels on its designated route from the city to the village. Each village has a Wi-Fienabled kiosk and when the bus is in range it picks up the data and buffers it. When the bus arrives in an area with internet access i.e. a HUB, the data is uploaded to the HUB and is collected when available. The data is delivered to the destination villages. In a report about empowering India, it was reported that in the DakNet project most messages were delivered within 6 hours [6]. This is clearly an indication of latency that is experienced. B. VANets In order for data to hop from one mule to another, the mules have to be connected to communicate. When the mules connect to each other they form an ad hoc network [7], furthermore if we assume the mules are partially vehicular, they form a type of vehicular ad hoc network (VANets) [8,9]. VANets are specialized mobile ad hoc networks (MANets). A MANet is a type of ad hoc network that changes locations and re-configures itself also the mules in the network are mobile, they use wireless connections to connect to various networks. VANets are distributed selforganizing communication networks built up from travelling vehicles and a subset of MANets. This means an algorithm designed for a MANet should be able to work for a VANet with a few changes to accommodate the differences between the networks. C. Optimal Relay Path Algorithm An algorithm that works well in a MANet is the optimal relay path algorithm (ORP) [7]. In brief, the algorithm uses the strong assumption that in an ad hoc network of mobile computers (data mules) the trajectory of each host is approximately known. ORP calculates the shortest path from one mule to another. It uses the current direction, speed and the path of the mule in order to calculate at what time the sender and receiver mule are in range. It also calculates if sending via an intermediate mule/s between the

sender and receiver will be faster. Some of the disadvantages with the ORP algorithm were that every mule in the network needed to know the current position and direction of all the mules in order to calculate the optimal path and there was an error incurred during the calculation of when to send a message. This meant that at the time that a sender mule sends a message to the intermediate/receiver mule; the mule may be out of range. III. PROPOSED DATA MULE INTERCOMMUNICATION ALGORITHM One of the research questions we try to answer is whether mule clustering would improve communication efficiency between isolated networks. This paper proposed to introduce clustering on top of the ORP algorithm. The idea is that when mules are within a certain range they form a temporary cluster for better and accurate exchange of data. A cluster is a subset of the network and mules in a cluster are all connected to each other. This means there should be a reliable path between the mules in the cluster. However, mules not in a cluster use the ORP algorithm to exchange data. One of the key factors that affect the design the algorithm for the multiple data mule inter-communication is the movements of the mule. The mule movements have a predictable pattern because the mule would be a bus or taxi equipped with a wireless data carrier. A bus or taxi has scheduled movements and travels the same route; however, the speed of the vehicle changes unpredictably. This is important in deriving our algorithm: Data Mule Intercommunication (DMI) algorithm, which is an integration of mules clustering and the ORP algorithm. The DMI algorithm uses the ORP algorithm for inter-cluster algorithm and simple clustering algorithm for intra-cluster communication. Figure 1 illustrates communication between mules and the clustering technique. Figure 1: Example of 5 mules, with a range of 50m, using the DMI algorithm to send messages. For example the mules at the top of the image belong to the same cluster set. The single mule on the far right does not belong to any cluster set.

mule. If a non-head mule wants to send a message, it sends the message to the head mule. IV. CONCLUSION The clustering integrated with the optimal relay algorithm is designed to improve the performance to the optimal relay algorithm. As mentioned, for inter-cluster communication, ORP algorithm is used. This means that the clusters are not used and the sender mule finds the optimal path and time to send the message to the intermediate/receiver mule. This is the new algorithm. V. FURTHER WORK In order for DMI to be used in the rural area we shall use mobility traces collected from a VANet mobility simulator that depicting vehicluar movement that allows for brief stops. Furthermore, we shall investigate the best size of a cluster. This is important for the mules to be as connected as often as posible. Finally, we shall test for how long the mule should stop at a network for and finally a comparison of the time taken to send a message using DMI vs. ORP. VI. REFERENCES [1] R.C. Shah, S. Roy, S. Jain, and W. Brunette, "Data MULEs: modeling a threetier architecture for sparse sensor networks," in Sensor Network Protocols and Applications, 2003. Proceedings of the First IEEE. 2003 IEEE International Workshop on, May 2003, pp. 30-41. [2] W Zhao, M Ammar, and E Zegura, "A Message Ferrying Approach for Data Delivery in Sparse MANets," in MobiHoc, Roppongi, Japan, 2004 May 24–26, pp. 187 - 192. [3] K Fall, "A Delay-Tolerant Network Architecture for Challenged Internets," in ACM SIGCOMM, 2003. [4] SearchingNetworking.com. (2011, May) SearchingNetworking.com. [Online]. http://searchnetworking.techtarget.com/definition/delay-tolerant-network [5] A Pentland, R Fletcher, and A. Hasson, "DakNet: Rethinking Connectivity in Developing Nations," Computer, vol. 37, no. 1, pp. 78-83, 2007. [6] R. P. Adler and M. Uppal, "m-Powering India: Mobile Communications for Inclusive Growth," The Aspen Institute and Aspen Institute India, Gurgaon, India, Third Annual Joint Roundtable on Communications Policy ISBN: 089843-492-0, 2008. [7] D Rus and Q Li, "Sending Messages to Mobile Users in Disconnected Ad-hoc Wireless Networks," in MobiCom '00: Proceedings of the 6th annual international conference on Mobile computing and networking, New York, NY, USA, 2000, pp. 44-55. [8] N Lagraa, M Yagoubi, and S Benkouider, "Localization technique in VANets using Clustering (LVC)," International Journal of Computer Science Issues, vol. 7, no. 4, pp. 10-16, July 2010. [9] M. Fiore, J. Harri, F. Filali, and C. Bonnet, "Vehiclar Mobility Simulation for VANets," Simulation Symposium, 2007. ANSS '07. 40th Annual , no. 1080-241X, pp. 301-309, March 2007. [10] R Agarwal and M Motwani, "Survey of clustering algorithms for MANET," International Journal on Computer Science and Engineering, vol. 1, no. 2, pp. 98-104, 2009. [11] Y. Chen, A. Liestman, and J. Liu, "Clustering Algorithms for Ad Hoc Wireless Networks," in DIALM '00 Proceedings of the 4th international workshop on Discrete algorithms and methods for mobile computing and communications, New York, NY, USA, 2000, pp. 54-63.

For intra-clustering we derived the clustering structure from a combination of several existing ad hoc wireless networks [10,11,12]. Mules within range of each other form a cluster. A cluster is formed if two mules are within range to each other and another mule may join the cluster if it is within range to one of the mules in the cluster. For each cluster there is a head mule and it knows the communication route to each mule from itself. The head mule is the most connected mule and has the shortest path to all the other mules in the cluster and each mule in the cluster knows the path to the head mule. All messages are sent from the head

[12] R Xu and R Wunsch, "Survey of Clustering Algorithms," IEEE tranaction on neural networks, vol. 16, no. 3, pp. 645 - 678, May 2005.

Xoluqobo Mkhwanazi received her undergraduate degree in 2008 from the University of Cape Town and is presently studying towards her Master of Science degree at the same institution. Her research interests include data mules, isolated wireless networks and clustering.