Browsing by Author "Douglas Reeves, Committee Chair"

Now showing 1 - 5 of 5

Design and Implementation of a Gnutella-based Reputation Management System
(2003-11-24) Murthy, Prashant; Munindar Singh, Committee Member; Peng Ning, Committee Member; Douglas Reeves, Committee Chair
Peer to peer (P2P) networks have introduced a new paradigm in content distribution. Such systems have shifted the paradigm from a client-server model into a client-client model. The tremendous success of such systems has proven that purely distributed search systems are feasible and that they may change the way we interact on the Internet. Most P2P protocols have been designed with minimum or no emphasis on security - Gnutella being one such open protocol standard. In this work, we focus on providing security over Gnutella by establishing trust between the entities (peers) in a P2P network using reputations and by ensuring integrity, authentication and non-repudiation of messages exchanged. Reputation systems collect, distribute and aggregate feedback about past behavior of the participants. Such systems help in establishing trust amongst strangers, detecting misbehaving nodes and isolating them. In this work, we analyze some existing reputation-based protocols in P2P networks. Among these protocols, we choose two approaches that are more specific and relevant to P2P networks. We compare these two protocols, namely, P2PRep and RCert in terms of security and performance. While P2PRep uses a broadcast polling mechanism and client-side storage to manage reputations, RCert uses unicast messages and server-side (local) storage of reputation content. Based on an analysis of the two approaches, we choose to enhance RCert. We identify the shortcomings and vulnerabilities of this protocol and propose an extension to RCert. We then provide the details of the design and implementation of our enhanced protocol - GTKgREP on Gtk-Gnutella, a unix-based Gnutella servent. We provide an assessment of the overheads associated with this protocol.
Improving Performance of Peer-to-peer Systems by Caching
(2005-01-09) Zhang, Qinghua; Douglas Reeves, Committee Chair; David Thuente, Committee Member; Khaled Harfoush, Committee Member
Recently, Peer-to-Peer (P2P) has attracted a great deal of interest in industry and research literature. P2P systems are application layer networks, in which logically distinct computing elements - peers, bear comparable roles and responsibilities. P2P enables peers to share resources in a distributed manner. Existing P2P systems work well but are inefficient with respect to information retrieval. Some measurements show that more than half of the current Internet traffic is P2P traffic. Some search methods currently used by P2P groups flood the network, thus consuming a lot of bandwidth. In addition, some P2P applications require some forms of global knowledge of peer resources. Caching is one way to improve the performance of any system that makes repetitive requests. This thesis proposes a selective query-forwarding scheme based on caching. This simple caching mechanism improves efficiency and scalability in information retrieval for P2P systems. Query processing is expedited by caching similar queries or replies, thus making searches more efficient. The performance of this caching-based search algorithm is evaluated and compared with two existing P2P search algorithms (flooding and Random Walk) in P2P file sharing systems. The simulation experiments are designed and performed based on some measurement and empirical data. The results show this caching-based scheme is an attractive technique for keyword based searching in P2P systems. In some cases it achieves 75\% query hits through caching. Its performance is also superior in that it consumes less bandwidth and takes less time to satisfy queries. Finally, this approach doesn't incur additional network traffic to develop knowledge on resource location and thus scales well with the size of the network.
NAMO NAMAHA: Network Assisted Multicast Overlay ConstructioN Algorithms for Mobile Ad Hoc Applications
(2003-12-19) Acharya, Mithun Puthige; Douglas Reeves, Committee Chair; Mladen Vouk, Committee Member; David Thuente, Committee Member
Group communication is the most important mode of communication in ad hoc networks, because of the collaborative nature of mobile ad hoc applications. In this light, an efficient and light weight multicast routing protocol is necessary. Presently the multicast routing is either done entirely at the network layer, or at the application layer as stateless overlay mulitcasting. Owing to the dynamic nature of ad hoc networks, the first method incurs a large signaling overhead due to frequent modification of routing tables and exchanging of session state information. The latter approach uses the underlying unicast routing to build multicast data distribution trees without maintaining session state information thereby trading efficiency for minimal messaging overhead. For small groups with constant bound on the number of multicast group members, the overlay schemes, apart from having a trivial signaling overhead, are also known to be far more efficient than the network layer schemes. But the existing overlay schemes do not completely exploit the 'knowledge' possessed by the network layer; they just use the unicast routing at the network layer. We believe that, even for larger groups, the overlay schemes can function with reasonable efficiency along with trivial signaling overhead if they intelligently use the network layer information. In this thesis, we propose a network assisted multicast routing scheme, NAMO NAMAHA, which primarily operates as an overlay while getting assistance from the network layer unicast routing protocol, CEDAR. The overlay algorithms dynamically build an approximate Steiner data distribution tree, adopting the (CHINS)_T (Cheapest Insertion Heuristic with Table) algorithm for the distributed implementation of the well known Takahashi-Matsuyama heuristic. The Steiner trees are incrementally built over a subgraph of core nodes, which form the approximate Minimum Dominating Set (MDS) over the network nodes. The core nodes get computed by a network layer heuristic using local data at that layer, and they provide useful information for the NAMO NAMAHA tree computation algorithms at the overlay layer. The main idea in this thesis is that if the construction of overlays is aided by some minimal but useful information from the network layer that is almost invariant, local and that which would incur constant memory overhead, efficient overlays can be constructed. This thesis presents the algorithms for the protocol NAMO NAMAHA, offers proof of correctness for the protocol and shows that the time and memory complexity of the algorithms in the protocol are either constant, or linear with the number of graph edges or nodes. We compare our work with the MCEDAR protocol in terms of the cost of the multicast data distribution trees, the number of messages exchanged in building them and the time and memory complexity of the algorithms involved. We choose MCEDAR since other multicast protocols for ad hoc networks are either network based which does not scale for large nubmer of nodes, or function as overlays designed only for small groups. When compared to MCEDAR, NAMO NAMAHA has a simpler join protocol implemented by our unique Unicast Trap algorithm that does not make use of acknowledgements. Unlike MCEDAR, the sender discovery messages are not propagated all over the network; they are restricted to regions where it is absolutely necessary. In NAMO NAMAHA, at any given time, a path exists between any multicast subscriber and the sender (which is approximately the best path possible) with very high probability. Such a guarantee cannot be given in MCEDAR. Furthermore because of the incremental Steiner tree construction, the resulting multicast data distribution tree has nearly the least total cost. Cost is not minimized in MCEDAR. These advantages are obtained in NAMO NAMAHA just by using extra messages during tree construction, whose number is well below the actual number of nodes in the multicast group. The time and memory complexity of the NAMO NAMAHA algorithms are in the same order as that in MCEDAR.
Network and Host Based Countermeasures against Large-scale Networked Compromised Systems or Malicious Software.
(2010-07-14) Park, Young Hee; Douglas Reeves, Committee Chair; Peng Ning, Committee Member; Ting Yu, Committee Member; Xuxian Jiang, Committee Member
Wolfsting: Extending Online Dynamic Malware Analysis Systems by Engaging Malware.
(2010-06-14) Mulukutla, Vikram; Douglas Reeves, Committee Chair; Xuxian Jiang, Committee Member; Peng Ning, Committee Member