Traditional P2P overlay networks 

Context in source publication

Context 1

... I NTRODUCTION Peer-To-Peer (P2P) and Mobile Ad hoc NETworks (MANETs) share the same key concepts of self-organization and distributing computing, and both aim to provide connectivity in a completely decentralized environment [1], [2]. Moreover, both lack central entities to which delegate the management and the coordination of the network and relay on a time-variant topology. In fact, in P2P networks the time- variability is due to joining/leaving peers, while in MANET ones it is due to both node mobility and propagation condition instability. Despite these similarities, the adoption of the P2P paradigm to disseminate and discover information in a MANET scenario rises to new and challenging problems [1], [3]. The main issue concerns the layer where they operate: P2Ps build and maintain overlay networks at the application-layer, assuming the presence of an underlying network routing which assures connectivity among nodes, while MANETs focus on providing a multi-hop wireless connectivity among nodes. This issue is a major problem in trying to couple a P2P overlay network over a MANET: in [4], [5] it has been proved that simply deploying P2P over MANETs may cause poor performances due to the lack of cooperation and communication between the two layers, causing so significant message overhead and redundancy. For these reasons, different cross-layer approaches have been presented and they can be classified according to the adopted solution for the resource discovery procedure. More specifically, in unstructured P2Ps, peers are unaware of the resources that neighboring peers in the overlay network maintain [6], [7]. So, they typically resolve search requests by means of flooding techniques and rely on resource replication to improve the lookup performance and reliability. Differently, in structured P2P networks peers have knowledge about the resources offered by overlay neighbors, usually by resorting to the Distributed Hash Table (DHT) paradigm and, therefore, the search requests are forwarded by means of unicast communications. Clearly, the scenarios where MANETS operate make unsuit- able both flooding and replication mechanisms, except for small networks and/or high joining/leaving peer rates. In the last years structured P2P networks have gained attention: EKTA [8] and DPSR [9] integrate a Pastry-like [10] structured P2P protocol with the DSR routing algorithm, while CROSS- Road [11] integrates a Pastry-like DHT over the OLSR routing algorithm, and VRR [12] proposes a routing algorithm which provides indirect routing by resorting to a Pastry-like structure too. All these techniques associate an identifier, namely a key, to each peer by means of an hash function and organize the keys in a ring structure. Since the identifiers are randomly assigned to peers, the P2P overlay topology is usually built independently from the physical one, and thus no relationship exists between overlay and physical proximity (Fig. 1). As shown in [13], [14], this implies that overlay hops can give rise to physical routes which are unnecessary long. Kademlia [15] shares several similarity with Indirect Tree-based Routing, in particular as regards to routing table maintenance. However the overlay and physical proximity are not fully related, since it resorts to a XOR-based distance, which cannot fully take into account the physical topology. MADPastry [16], [17] integrates the Pastry protocol with the AODV routing algorithm and tries to overcome this issue by resorting to clustering. However, the overlay and physical proximity are in someway related only for inter-cluster communications. In [18], it is proposed to associate location-dependent identifiers to nodes with a distribute procedure and to organize node in a tree-based overlay structure. In this paper, according to [18], we give a contribution toward the structured P2P approach presenting a DHT-based routing protocol, namely Indirect Tree-based Routing (ITR), which integrates both traditional direct routing and indirect key-based routing at the network layer. Indirect Tree-based Routing extends the Augmented Tree-based Routing (ATR) [19], a hierarchical multi-path routing protocol for scalable ad-hoc networks, by providing fully functional P2P services. Like [19], we resort to an augmented tree-based structure, in order to assure that the logical and the physical proximity agree, as shown in Fig. 2. For both direct and indirect routing, each node maintains a unique routing table which stores only physical 1-hop neighbors, i.e. only peers with which the node can communicate at the link layer. As result, each overlay hop consists of only one physical hop. To test the effectiveness of our proposal, numerical simulations on 802.11 technology have been carried out across a wide range of environments and workloads. It is worthwhile to underline that ITR can be accommodated with slight modifications to operate over any link layer technology and, moreover, it does not require any change in both transport and application layers. The outline of the paper is the following: Section II presents the design and implementation details of ITR, whereas Section III presents the performance evaluation. Finally, in the last section conclusion and open problems are drawn. II. I NDIRECT T REE - BASED R OUTING As mentioned before, Indirect Tree-based Routing extends the Augmented Tree-based Routing (ATR) by providing fully functional P2P services. Both assign location-dependent identifiers, namely strings of l bits, to peers by means of a distribute procedure and of locally broadcasted hello packets. The peer identifier space can be represented as a complete binary tree of l levels, that is a binary tree in which every vertex has zero or two children and all leaves are at the same level (Fig. 2-a). In the tree structure, each leaf is associated with a peer identifier, and a inner vertex of level k , namely a level-k subtree , represents a set of leaves (that is a set of peer identifiers) sharing a prefix of l − k bits. For example, with reference to Fig. 2-a, the vertex with the label 01X is a level- 1 subtree and represents the leaves 010 and 011 . Let us define as level-k sibling of a leaf as the level- k subtree which shares the same parent with the level- k subtree the leaf belongs to. Referring to the previous example, the vertex with the label 1XX is the level- 2 sibling of the address 000 . Indirect Tree-based Routing performs the whole routing resorting to an iterative procedure which explores the topological meaning of the node identifiers with a hierarchical form of multi-path proactive distance-vector routing. Like ATR, each node stores a routing table with l sections, one for each sibling, and the k -th section stores the physical 1-hop neighbor peers which can forward a packet towards peers whose location-dependent identifiers belong to the level- k sibling. With reference to the topology depicted in Fig. 3 where the location identifiers are 5 bit long, we suppose that the node with identifier 10000 has to communicate with the node with identifier 00000 . Since 00000 belongs to the level- 4 sibling of identifier 10000 , the source will forward a packet to the physical neighbor with identifier 01000 , according to its routing table shown in Fig. 4 (further details on the routing table maintenance could be found in [19]). From an operational point of view, Indirect Tree-based Routing performs like traditional P2P systems: namely, when a node stores a resource, it sends periodically a ...