Optimally Efficient Prefix Search and Multicast in Structured P2P Networks

Technical Report TUD-CS-2008-103, Technische Universit¨at Darmstadt, 08.08.2008 Optimally Efficient Prefix Search and Multicast in Structured P2P Networks Technical Report No. TUD-CS-2008-103 Telecooperation Report No. TK-01/08, The Technical Reports Series of the TK Research Division, Technische Universit¨at Darmstadt D. Bradlera∗, J. Kangasharjub, M. M¨uhlh¨ausera aTelecooperation Group, University of Technology Darmstadt, Darmstadt, Germany bDepartment of Computer Science, University of Helsinki, Finnland Abstract Searching in P2P networks is fundamental to all overlay networks. P2P networks based on Distributed Hash Tables (DHT) are optimized for single key lookups, whereas unstruc- tured networks offer more complex queries at the cost of increased traffic and uncertain success rates. Our Distributed Tree Construction (DTC) approach enables structured P2P networks to perform prefix search, range queries, and multicast in an optimal way. It achieves this by creating a spanning tree over the peers in the search area, using only information available locally on each peer. Because DTC creates a spanning tree, it can query all the peers in the search area with a minimal number of messages. Furthermore, we show that the tree depth has the same upper bound as a regular DHT lookup which in turn guarantees fast and responsive runtime behavior. By placing objects with a region quadtree, we can perform a prefix search or a range query in a freely selectable area of the DHT. Our DTC algorithm is DHT-agnostic and works with most existing DHTs. We evaluate the performance of DTC over several DHTs by comparing the performance to existing application-level multicast solutions, we show that DTC sends 30–250% fewer messages than common solutions. ∗this research is funded by DFG arXiv:0808.1207v1 [cs.NI] 8 Aug 2008 2 D. Bradler, J. Kangasharju, M. M¨uhlh¨auser 1. Introduction Peer-to-peer networks are typically divided into unstructured and structured networks, depending on how the overlay is constructed and how content is placed in the network. In an unstructured network, nodes are free to choose their overlay neighbors and are free to offer any content they want. In contrast, in a structured network, the structuring mechanism (typically one or more hash functions) uniquely determine the location of a peer in the overlay and its neighbors, as well as the placement of content on peers. Both these kinds of networks have their strengths and weaknesses. Unstructured net- works are based on searching for content, which allows us to use complex queries for determining which objects match which requests. However, this comes at the expense of having to flood the search query through the network which causes significant amount of network traffic. The original Gnutella network was a completely flat overlay which relied on locating objects by flooding the network up to a given time-to-live (TTL). Newer un- structured networks, like Kazaa and eDonkey, build a two-tier hierarchy with ordinary peers and superpeers. Ordinary peers connect to one superpeer and the superpeers build a Gnutella-like overlay between them. Although such a hierarchical network scales better than the plain Gnutella network, we still cannot guarantee that an object can be found and the amount of network traffic caused by flooding between the superpeers can still be high. A structured network, on the other hand, clearly defines the overlay structure and object placement through hash functions. The main advantage is that it allows for very efficient key-value lookups, similar to traditional hash tables. However, because the content is placed with hash functions, real search queries are not feasible in DHTs. For example, it is not feasible to query a DHT for all objects whose name begins with “Foo”. This would usually require asking every peer whether it has any matching objects. We tackle the problem of implementing a prefix search on a structured P2P network. Our algorithm selects a subset of the overlay network, and creates a spanning tree for that subset rooted at any of the peers in that set. Typical applications which benefit from prefix search are all systems which need to deal with structured data. Structured data is very commonly used in many applications. For example storing data based on geographical location (e.g., tourist information) or any kind of classification systems lend themselves readily to our algorithm. The more attributes are available in the user provided search term (i.e., the more precise the user’s query), the less nodes will be queried. In the case of a user performing a very precise search, the query will reduce itself to a standard DHT key lookup. Because we create a spanning tree, our solution is not limited to implementing prefix search; other applications, such as multicast and broadcast, can also be implemented with our algorithm. Multicast (and also broadcast) is extremely useful in cases where the DHT is built according to some specific criteria (as op

…(Full text truncated)…

This content is AI-processed based on ArXiv data.