Reliable Mobile Ad Hoc P2P Data Sharing

Transcription

1 Reliable Mobile Ad Hoc P2P Data Sharing Mee Young Sung 1, Jong Hyuk Lee 1, Jong-Seung Park 1, Seung Sik Choi 1, and Sungtek Kahng 2 1 Department of Computer Science & Engineering, University of Incheon {mysung, hyuki, jong, sschoi}@incheon.ac.kr 2 Department of Information & Telecommunication Engineering, University of Incheon s-kahng@incheon.ac.kr Abstract. We developed a reliable mobile Peer-to-Peer (P2P) data sharing system and performed some experiments for verifying our routing scheme using our real testbed. Our method for guaranteeing reliable P2P data sharing mainly deals with the problem of disconnection due to the sudden disappearance of one or more nodes involved in the active transmission route. The routing scheme of our system allows for reliable transmission of data via mobile devices. It does this using our reconnection mechanism that finds emergency routes in instances of abrupt disconnection using our scanning table and lookup table. The experiments lead us to conclude that our communication method is effective and assures reliable P2P data sharing over mobile ad hoc networks. In addition, our system allows for finding directly connectable devices at the overlay networks, therefore the application program can distinguish whether the target node is directly accessible or not with the aid of our scanning table. 1 Introduction Mobile ad hoc networks (MANETs) are envisioned to become key components in 4G (fourth-generation) wireless network architecture. MANETs inherit common characteristics found in most wireless networks, and add characteristics specific to ad hoc networking, such as wireless, ad-hoc-based, autonomous, infrastructureless, multihop routing, and mobility [1]. Peer-to-Peer (P2P) networks act independent from the underlying network infrastructure. The TCP/IP only provides a communication platform protocol and does not supply information about the location of requested content or participants. P2P nodes independently maintain P2P networks and show characteristics of a free infrastructure and decentralized network [2]. P2P as well as mobile ad hoc networks follow the same idea of creating a network without the help of central entities. A combination of both, produces synergies and creates new possibilities, but also poses several problems. The mobile ad hoc P2P paradigm attempts to create open and collaborative networks of a very diverse functionality and nature. Such functionality extends from the most popular data sharing protocols, like Gnutella, up to P2P instant messengers and chat applications, etc. However, regardless of its various applications, P2P P. Lorenz and P. Dini (Eds.): ICN 2005, LNCS 3421, pp , Springer-Verlag Berlin Heidelberg 2005

2 Reliable Mobile Ad Hoc P2P Data Sharing 773 networking always involves unstable transmission due to the mobility of some nodes which were engaged in the transmission. This means that P2P computing needs to assure stable data transmission until the successful termination of services. A mobile ad hoc P2P data sharing system autonomously connects a dynamically changing set of nodes to allow the discovery and transfer of information among them. Together, a mobile ad hoc P2P data sharing system exhibits common characteristics that come with having rapidly and unpredictably changing users and not set infrastructure. Moreover, mobile devices are constrained by battery energy and computation capabilities, and communicate via low bandwidth wireless links [3]. Therefore, it is important to guarantee the stable and efficient transmission of information even though some service nodes suddenly disappear because of the movement of users or power shortage of the battery. In this paper, we propose a reconnection mechanism for finding the emergency routes that guarantee reliable P2P data sharing over mobile ad hoc networks. By the term emergency route we define a new route which can substitute the broken route of the current service. The paper is organized as follows: In the following section we briefly present the related work. Subsequently, we describe our ongoing development of mobile ad hoc P2P data sharing system. In Section 4, we discuss our proposed routing scheme for mobile ad hoc P2P data sharing application. In Section 5, we describe our experimental setup and discuss the results obtained in terms of their reliability as well as the throughput (reconnection or transmission time). Finally, we conclude in the last section. 2 Related Work The primary objective of an ad hoc network routing protocol is the correct and efficient route establishment between a pair of nodes so that messages can be delivered reliably and in a timely manner. In general, ad hoc network routing protocols can be divided into two broad categories: proactive routing protocols and reactive on-demand routing protocols [4]. The most representative proactive protocol is DSDV (Destination-Sequenced Distance-Vector). The most representative reactive protocol is AODV (Ad-hoc on-demand Distance Vector). A P2P network structure can be classified in two ways: a centralized structure and a decentralized structure. A decentralized networking structure can in turn be classified again as structured or unstructured. The unstructured architecture does not manage any linkage information of peers and the locations of files until explicit requests are made. Peers update their networking information only when it is requested. The structured architecture always retains the information about a network whenever a node joins and leaves. An example of a structured type is Chord [5] and an example of an unstructured type is ORION (Optimized Routing Independent Overlay Network) [6]. ORION is a file discovery algorithm based on keyword searching. ORION combines application-layer query processing with the network layer process of route discovery. Chord is based on the DHT (Distributed Hash Table)

3 774 M.Y. Sung et al. algorithm which uses the distributed hash values of nodes and keys. Chord solves the problem of storing the lookup data in a decentralized manner. Note that ORION, Chord, and our work can be classified as reactive routing protocols. From a structuring point of view, our work can be classified as both structured and unstructured (The scanning table of our work is composed in structured manner and the lookup table of our work is composed in unstructured manner). 3 Mobile Ad Hoc P2P Data Sharing System Our mobile ad hoc P2P data sharing system allows users to share any information on files. What distinguishes our system from others is that it provides an automatic configuration of ad hoc networking modes without manual setting or execution of software by users. It also automatically generates the private IPs without any user intervention. Our system starts with the establishment of ESSID (Extended Service Set IDentifier). If one of the peers is started, any other peers with the same ESSID will be started and discover other mobile devices. Once peers are connected to each other, they share information about each other s routing tables and then update their own routing tables. Our system is implemented using the embedded Linux development kits. On the development kit, the embedded Linux kernel is installed and we used airo.o as a PCMCIA wireless LAN card, libiw.so.26 as our wireless network library, and Qt library ver.3.07 for user interfaces. Main components of our system are as follows: A Discovery Engine: It recognizes BSSID (Basic Service Set IDentifier) or ESSID (Extended Service Set IDentifier), translates the MAC addresses into IPs, and creates the scanning table and the lookup table which will be discussed in the following section. A Reconnection Engine: It is responsible for connection management and it provides the reestablishment of alternate routes or queries. A Transmission Engine (composed of a Packet Manager, a Control Socket, and a Data Socket): It takes charge of data transmission and it is implemented using Linux sockets. The Packet Manager controls the routing, and the actual file transmission is processed by Data Socket. A Management Module: It supports the QoS based data buffering for efficient data transmission. Presentation Module: It corresponds to the GUI (Graphical User Interface) part of the system and it is implemented using the Qt library. Any manipulation events entered through the GUI are processed by the Controller in Management module. 4 Routing Scheme Many routing mechanisms and communication protocols are studied in the field of mobile ad hoc networks [7], [8], [9]. In general, a mobile ad hoc P2P data sharing

4 Reliable Mobile Ad Hoc P2P Data Sharing 775 system consists of a search mechanism to transmit queries (lookup) and search results, and a transfer protocol to download files matching a query. Recent studies on mobile ad hoc P2P computing are mainly focused on the efficient search of nodes that store the desired data item. One of the pioneering works is Chord which aims for a good hit ratio. One other work is ORION, which searches mobile devices that hold target information whilst using the least bandwidth. Both discovery algorithms combine application layer query processing with the network layer process of route discovery. Our routing algorithm is composed of the following two mechanisms: An efficient discovery mechanism: This mechanism discovers the devices first and searches routing information for locating the target item. It is provided by our discovery engine which creates the routing table which is composed of a scanning table and a lookup table. Once the target is found, our reliable transmission algorithm starts to process. A reliable transfer mechanism: This takes charge of reliable communication thus guaranteeing the successful completion of requested services. It includes a reconnection engine in cases where a sudden disappearance of service nodes occurs because of the movement of users or a shortage of battery energy. Our reconnection engine is based on the combined usage of the following two reconnection algorithms: A dynamic-alternate algorithm: It tries to find an alternate route in searching the lookup table. The lookup table of our system is similar to the file routing table of ORION. However, our lookup table contains information about the final destination of the target item, while ORION only keeps information about the node who knows the location of the target item. Our dynamic-alternate algorithm allows for finding the alternate route which is directly accessible. A dynamic-recruiting algorithm: It first selects one other neighbor node from amongst the nodes on the scanning table (which holds the directly accessible nodes in the active transmission range of a peer) and it retries until it finds a new route for accessing the target item. Whenever a peer is connected to another neighbor node, their lookup tables will be updated by sharing each other s information. 4.1 Automatic Configuration of Ad Hoc Networks Our system allows for automatic configuration of ad hoc networks. Once a mobile device starts, the mode of each mobile device within an active transmission range has to be changed into the ad hoc mode and each device is set with the same ESSID (Extended Service Set IDentifier). Then the mobile device becomes accessible within the active ad hoc network. However, the application program cannot connect to mobile devices without IP addresses. Therefore, we need to provide a private IP for each mobile device. Our system automatically allocates private IPs to mobile devices in an active transmission range at the moment of initial ad hoc network configuration. Our algorithm is designed for overlay networks. It assumes that the target information is within two hops and constructs the routing table using a scanning table

5 776 M.Y. Sung et al. and a lookup table. The scanning table is constructed when the ad hoc network is initially configured and it contains mobile nodes which are directly connectable. It is possible to find directly connectable devices at the level of link layer using MANET routing algorithms such as AODV. However, the application program still needs to know the IPs for data transmission. Therefore, we build the scanning table containing directly accessible IPs and the corresponding flags at the overlay network level. We distinguish three values of flags: 0 means a node itself, 1 means the directly connectable node, and 2 means the node out of active area. Whenever a node appears within active transmission range, it can define its own flag. The node with flag 2 is not directly accessible. Only the nodes with flags 0 or 1 remain in the scanning table. Once node A starts to work, it broadcasts a message to discover its neighbors. If node A receives no answers, it takes an IP and sets its flag to 0. If node B comes in the active area of node A and broadcasts a message for discovering then it receives a response (the content of the scanning table) from node A. In this case, node B sets its IP, which is not the same as node A s IP and its flag to 0, and keeps the reply from node A. If node C enters the active area of node B which is not accessible from node A. Node C receives the content of the scanning table from node B and comes to know about the existence of nodes B and A. Node C sets its IP and its flag to 0 and sets node B s flag to 1 because it is accessible. As node C cannot receive any response from node A, it sets node A s flag to Scenarios of Discovery and Transmission To illustrate the operation of our discovery and transmission algorithm, we should consider the six-node scenario shown in Figure 1. Nodes are connected by an arrow and the rectangles near the nodes show parts of the lookup table. The large circle corresponds to the boundary of an active wireless transmission range. In Figure 1, node A can find 5 nodes (B~F) which are bound with the same ESSID. Once a connection is made between two nodes, they update their lookup tables by referencing the lookup table of the counter part. Fig. 1. Six-node Scenario Fig. 2. Scenario 1 Fig. 3. Scenario 2 Our routing table is composed of two parts: one is the scanning table and the other is the lookup table. The scanning table contains all of the mobile devices which are

6 Reliable Mobile Ad Hoc P2P Data Sharing 777 found through the scanning broadcasting at the link layer. The scanning broadcasting is implemented using libiw.so.26, which is a C-language library at the MAC layer of IEEE The lookup table is created at the application level and it contains pairs of the target item and the node which holds the target item. There are two types of transmission method. One is to get information in multiple parallel channels from many peers. Another is to get the information from only one peer. To get information in parallel is effective in wired networks. However, it is not desirable to use in wireless networks due to the lack of transmission capability of mobile nodes and the difficulties to allocate a large bandwidth for every channel. Therefore, we limit the number of peers that are connected at the same time to 10 peers and the number of peers transmitting data in parallel to 3. As mentioned before, our discovery and transmission algorithms assume that a connection between two nodes implies sharing and updating of each other s lookup tables. In Scenario 1 as shown in Figure 2, node A is connected to node B which is not connected to any node. Node A is downloading the target item from node B. In this situation, if node B is removed in the middle of transmission, node A has to try to connect to other nodes in the scanning table because there is no other information concerning the location of the target item in A s lookup table. Scenario 2 illustrated in Figure 3 corresponds to a situation where node A is connected to node B which is connected to another node C. Node A shares the lookup table of node B and already knows that C contains the target item. If node B disappeared in the middle of transmission, node A could directly connect to node C for downloading the target item. In Scenario 3 demonstrated in Figure 4, node A is connected to node B which is connected to many other nodes. Node A shares the lookup table of node B which includes information about the node that responded first to the lookup query. Assume that node E responded first in this example, and if node B is removed in the middle of transmission, node A can directly connect to node E for downloading the target item. Fig. 4. Scenario 3 Fig. 5. Scenarios 4 Scenario 4 presented in Figure 5 is similar to Scenario 3. Node A is connected to node B which is connected to many other nodes. Node A shares the lookup table of node B which includes information about the node that responded first to the lookup query for the target item. Assume that node F responded first. Node F is within the

7 778 M.Y. Sung et al. active transmission range of B, however it is out of the transmission range of A. In this situation, if node B disappeared in the middle of transmission, node A has to download the target item from the destination node (node F in this example) via the relay node B. 4.3 What Distinguishes Our Algorithm Although some features of our routing algorithm are similar to those of ORION, our algorithm distinguishes itself from ORION in the following ways: Our method for creating the routing table is different from that of ORION. ORION creates a response routing table (which contains routing information as well as lookup results), instead we separate the routing table into two parts: a scanning table and a lookup table. Processing lookup queries require more bandwidth than creating a simple routing table, therefore constructing the lookup table is more time consuming than constructing a scanning table for a direct connection and the lookup table for the location of items. In addition, our lookup table contains information about the final destination of the target item, while ORION only keeps the information of the relay node who knows the location of target item. Therefore, our dynamic-alternate algorithm allows for discovery and connection to a directly accessible emergency route. Our relaying method is also different. Our system provides direct connections to nodes as often as possible and tries to avoid connection to them indirectly. The efficiency of our system comes from the scanning table which contains directly accessible nodes. If the direct connection is not possible, the lookup table is searched for a relayed connection. 5 Experiments Using the real testbed based on our mobile ad hoc P2P data sharing system, we undertook some experiments for validating the effectiveness of our reconnection mechanism for assuring the successful accomplishment of the current service. We can make diverse scenarios of experiments according to the different numbers of intermediate nodes in the route for on-going transmission service. However, any scenario of experiments can be classified in the following two ways: The disappearance of a node in the current transmission route and the system knows the alternate route (for example, let each of A, B, C, and D correspond to a node in an ad hoc network, A wants D s information and A already knows two routes to reach D; the first one is from A to B then from B to D, and the second one is from A to C then from C to D, while communicating through the first route, B is removed and tries to reconnect to D). In this case, our system applies a dynamic-alternate algorithm for reconnection. The disappearance of a node in the current transmission route and the system does not know about the alternate route (for example, A wants C s information and A knows B and B knows C, while A communicating with C through B, B is

8 Reliable Mobile Ad Hoc P2P Data Sharing 779 removed and tries to reconnect to node C). In this case, our system uses a dynamic-recruiting algorithm. According to four diverse scenarios presented above, we calculated the reconnection time for the target information and the average transfer rate for downloading the target information. For the average transfer rate, we compared the performance of direct transmission (one-to-one, without relays) for mobile ad hoc P2P data sharing of a 1.5 Mbytes mp3 file and that of relayed transmission. We measured the throughput using the following formulas: SearchTime = ST, TransmissionTime = T TransmissionCapacity = TC TC Throughput = ST + T The experiment on the reconnection time showed that the time for reconnection increases in proportion to the number of shared files (in average, it takes 1.6 msec more per file). When measuring the performance of transmission, the average direct transfer rate (169 Kbytes/second) is approximately twice that of relayed transmission (83 Kbytes/second) as illustrated in Figure 6 (a) and Figure 6 (b). ª ª ª ª ª ª Fig. 6. Average Transfer Rates: (a) Average Transfer rate of direct transmission for downloading a 1.5Mbyte-mp3 file, (b) Average Transfer rate of relayed transmission for downloading a 1.5Mbyte-mp3 file 6 Conclusion We believe that the most important factor to consider when designing mobile ad hoc applications is to guarantee the completeness of services. Therefore, we developed a mobile P2P data sharing system which assures reliable and stable data transmission even though some service nodes suddenly disappear because of the movement of users or the shortage of power. Our system allows for configuring a MANET without intervention of users by allocating the IPs automatically. The routing scheme of our system selectively applies two reconnection methods; one is the dynamic-alternate algorithm and the other is the dynamic-recruiting algorithm, for finding emergency routes in the case of the abrupt disconnection. Our routing algorithm needs to create a

9 780 M.Y. Sung et al. scanning table (which contains the directly connectable nodes) as well as a lookup table (which contains the information about the location of target data) for augmenting reliability in data sharing services. Creating those tables requires some overheads, however it is necessary to eliminate unnecessary transmission relays. Through some experiments, we confirmed that our routing scheme for assuring reliable P2P data sharing over mobile ad hoc networks is effective. Acknowledgement. This work was supported by the Korea Science and Engineering Foundation (KOSEF) through the Multimedia Research Center at the University of Incheon. References 1. J.J.-N. Liu and I. Chlamtac, Mobile Ad hoc Networking with a View of 4G Wireless: Imperatives and Challeges, Mobile Ad Hoc Networking, Chapter 1, S. Basagni, M. Conti, S. Giordano, and I. Stojmenovic, Eds., Wiley, 2004, pp R. Schollmeier, I. Gruber, and F. Niethammer, Protocol for Peer-to-Peer Networking in Mobile Environments, Proceedings on the IEEE International Conference on Computer Communications and Networks (ICCCN) 2003, pp , A. Duran and C.C. Shen, Mobile Ad hoc P2P File Sharing, Proceedings on the IEEE Wireless Communications and Networking Conference (WCNC) 2004, vol.1., pp , E.M. Royer and C.-K. Toh, A Review of Current Routing Protocols for ad-hoc Mobile Wireless Networks, IEEE Personal Communications, April I. Stoica, R. Morris, D. Lieben-Nowell, D.R. Karger, M.F. Kaashoek, F. Dabek, and H. Balakrishnan, Chord: a Scalable Peer-to-Peer Lookup Protocol for Inrternet Applications, IEEE/ACM Transactions on Networking, vol.11, No.1, pp.17-32, February A. Klemm, C. Lindemann, and O. Waldhorst, A Special-Purpose Peer-to-Peer File Sharing System for Mobile Ad hoc Networks, Proceedings on the Vehicular Technology Conference (VTC) 2003, vol.4, pp , October 6-9, G. Anastasi, M. Conti, and E. Gregori, IEEE Ad Ho Networks: Protocols, Performancem, and Open Issues, Mobile Ad Hoc Networking, Chapter 3, S. Basagni, M. Conti, S. Giordano, and I. Stojmenovic, Eds., Wiley, 2004, pp J.P. Macker and M. Scott Corson Mobile Ad Hoc Networks (MANETs): Routing Technology for Dynamic Wireless Networking, Mobile Ad Hoc Networking, Chapter 9, S. Basagni, M. Conti, S. Giordano, and I. Stojmenovic, Eds., Wiley, 2004, pp M. Conti, E. Gregori, and G. Turi, Towards Scalable P2P Computing for Mobile Ad Hoc Networks, Proceedings on the Second IEEE Annual Conference on Pervasive Computing and Communications Workshops (PERCOMW 04), 2004.