Christian von der Weth

A unifying framework for behavior-based trust models

Trust models have been touted to facilitate cooperation among unknown entities Existing behavior-based trust models typically include a fixed evaluation scheme to derive the trustworthiness of an entity from knowledge about its behavior in previous interactions This paper in turn proposes a framework for behavior-based trust models for open environments with the following distinctive characteristic Based on a relational representation of behavior-specific knowledge, we propose a trust-policy algebra allowing for the specification of a wide range of trust-evaluation schemes A key observation is that the evaluation of the standing of an entity in the network of peers requires centrality indices, and we propose a first-class operator of our algebra for computation of centrality measures This paper concludes with some preliminary performance experiments that confirm the viability of our approach.

COBS: Realizing Decentralized Infrastructure for Collaborative Browsing and Search

Finding relevant and reliable information on the web is a non-trivial task. While internet search engines do find correct web pages with respect to a set of keywords, they often cannot ensure the relevance or reliability of their content. An emerging trend is to harness internet users in the spirit of Web 2.0, to discern and personalize relevant and reliable information. Users collaboratively search or browse for information, either directly by communicating or indirectly by adding meta information (e.g., tags) to web pages. While gaining much popularity, such approaches are bound to specific service providers, or the Web 2.0 sites providing the necessary features, and the knowledge so generated is also confined to, and subject to the whims and censorship of such providers. To overcome these limitations we introduce COBS, a browser-centric knowledge repository which enjoys the inherent openness (similar to Wikipedia) while aiming to provide end-users the freedom of personalization and privacy by adopting an eventually hybrid/p2p back-end. In this paper we first present the COBS front-end, a browser add-on that enables users to tag, rate or comment arbitrary web pages and to socialize with others in both a synchronous and asynchronous manner. We then discuss how a decentralized back-end can be realized. While Distributed Hash Tables (DHTs) are the most natural choice, and despite a decade of research on DHT designs, we encounter several, some small, while others more fundamental shortcomings that need to be surmounted in order to realize an efficient, scalable and reliable decentralized back-end for COBS. To that end, we outline various design alternatives and discuss qualitatively (and quantitatively, when possible) their (dis-)advantages. We believe that the objectives of COBS are ambitious, posing significant challenges for distributed systems, middleware and distributed data-analytics research, even while building on the existing momentum. Based on experiences from our ongoing work on COBS, we outline these systems research issues in this position paper.

Towards truthful feedback in p2p data structures

Peer-to-Peer data structures (P2P data structures) let a large number of anonymous peers share the data-management workload A common assumption behind such systems is that peers behave cooperatively But as with many distributed systems where participation is voluntary, and the participants are not clearly observable, unreliable behavior is the dominant strategy This calls for reputation systems that help peers choose reliable peers to interact with However, if peers exchange feedback on experiences with other peers, spoof feedback becomes possible, compromising the reputation system In this paper we propose and evaluate measures against spoof feedback in P2P data structures While others have investigated mechanisms for truthtelling recently, we are not aware of any studies in P2P environments The problem is more difficult in our context because detecting unreliable peers is more difficult as well On the other hand, a peer can observe the utility of feedback obtained from other peers, and our approach takes advantage of this To assess the effectiveness of our approach, we have conducted extensive analytical and experimental evaluations As a result, truthful feedback tends to have a much higher weight than spoof feedback, and collaboration attacks are difficult to carry out under our approach.

Strategic Provider Selection in a Policy-Based Helping Scenario

In service-oriented computing, software agents interact by requesting and providing services. Since providing a service incurs cost, uncooperative behavior dominates in the absence of an incentive mechanism. An economic model that describes interactions between individuals is the Helping Game. There, pairs of requester and provider are randomly matched. In various real-world applications in turn, several providers offer similar services, and requesters have a choice of providers. The rationale behind strategic provider selection is to choose the provider that is most likely to perform the task as desired. The results from existing studies of the Helping Game are not directly applicable to settings with provider selection. To analyze how strategic provider selection affects the efficiency of enterprise systems, we have designed and carried out an experimental study. Our results show that cooperative participants receive significantly more requests than uncooperative ones, making cooperation expensive. We conclude that system designers must incentivize requesters to balance their tasks between providers.

Optimizing Multiple Centrality Computations for Reputation Systems

In open environments, deciding if an individual is trustworthy, based on his past behavior, is fundamentally important. To accomplish this, centrality in a so-called feedback graph is often used as a trust measure. The nodes of this graph represent the individuals, and an edge represents feedback that evaluates a past interaction. In the open environments envisioned where individuals can specify for themselves of how to derive their trust in others, we observe that several centrality computations take place at the same time. With centrality computation being an expensive operation, performance is an important issue. While techniques for the optimization of a single centrality computation exist, little attention so far has gone into the computation of several centrality measures in combination. In this paper, we investigate how to compute several centrality measures at the same time efficiently. We propose two new optimization techniques and demonstrate their usefulness experimentally both on synthetic and on real-world data sets.

Fault-tolerant query processing in structured P2P-systems

Recently, a number of query processors has been proposed for the evaluation of relational queries in structured P2P systems. However, as these approaches do not consider peer or link failures, they cannot be deployed without extensions for real-world applications. We show that typical failures in structured P2P systems can have an unpredictable impact on the correctness of the result. In particular stateful operators that store intermediate results on peers, e.g., the distributed hash join, must protect such results against failures. Although many replication schemes for P2P systems exist, they cannot replicate operator states while the query is processed. In this paper we propose an in-query replication scheme which replicates the state of an operator among the neighbors of the processing peer. Our analytical evaluation shows that the network overhead of the in-query replication is in O(1) regarding network size, i.e., our scheme is scalable. We have carried out an extensive experimental evaluation using simulations as well as a PlanetLab deployment. It confirms the effectiveness and the efficiency of the in-query replication scheme and shows the effectiveness of the routing extension in networks of varying reliability.

Optimization Techniques for Multiple Centrality Computations

A broad range of data has a graph structure, such as the Web link structure, online social networks, or online communities whose members rate each other (reputation systems) or rate items (recommender systems). In these contexts, a common task is to identify important vertices in the graph, e.g., influential users in a social network or trustworthy users in a reputation system, by means of centrality measures. In such scenarios, several centrality computations take place at the same time, as we will explain. With centrality computation being expensive, performance is crucial. While optimization techniques for single centrality computations exist, little attention so far has gone into the computation of several centrality measures in combination. In this paper, we investigate how to efficiently compute several centrality measures at a time. We propose two new optimization techniques and demonstrate their usefulness both theoretically as well as experimentally on synthetic and on real-world data sets.