Search

Due to large volumes of data available online, techniques such as document classification and clustering are required for organization, analysis and management of data. Similarity-based Clustering (SBC) is used by many existing systems for filtering information. Decentralized gossip-based overlays offer a simple, robust and scalable solution to SBC clustering. Convergence and communication complexity are the two key areas of concern when SBC is implemented using these overlays. Convergence guarantees accurate clustering but costs bandwidth because these systems rely on message passing to achieve convergence. In this work, we address the long tail problem, experienced by low in-degree nodes in a long tailed similarity distribution, such as power-law distribution and propose a new SBC approach, Optimum Benefit Protocol (OBP), that converges more rapidly than existing approaches and reduces the long tail. The proposed protocol only sends messages that could benefit the receiver, reducing bandwidth to one fifth of the default. Our protocol obtains at least 90% convergence for a 900 node network, starting in a random configuration, in less than 10 cycles, for all observed experiments. We used real world distributions from Yahoo, Movielens, and Epinion datasets, for experimentation.

The K-Means algorithm for cluster analysis is one of the most influential and popular data mining methods. Its straightforward parallel formulation is well suited for distributed memory systems with reliable interconnection networks, such as massively parallel processors and clusters of workstations. However, in large-scale geographically distributed systems the straightforward parallel algorithm can be rendered useless by a single communication failure or high latency in communication paths. The lack of scalable and fault tolerant global communication and synchronisation methods in large-scale systems has hindered the adoption of the K-Means algorithm for applications in large networked systems such as wireless sensor networks, peer-to-peer systems and mobile ad hoc networks. This work proposes a fully distributed K-Means algorithm (EpidemicK-Means) which does not require global communication and is intrinsically fault tolerant. The proposed distributed K-Means algorithm provides a clustering solution which can approximate the solution of an ideal centralised algorithm over the aggregated data as closely as desired. A comparative performance analysis is carried out against the state of the art sampling methods and shows that the proposed method overcomes the limitations of the sampling-based approaches for skewed clusters distributions. The experimental analysis confirms that the proposed algorithm is very accurate and fault tolerant under unreliable network conditions (message loss and node failures) and is suitable for asynchronous networks of very large and extreme scale.

Peer-to-peer Desktop Grids provide integrated computational resources by leveraging autonomous desktop computers located at the edge of the Internet to offer high computing power. The arbitrary arrival and serving rates of tasks on peers impedes the high throughput in large-scale P2P Grids. We propose a novel autonomous resource allocation scheme, which can maximize the throughput of self-organizing P2P Grid systems. Our design possesses three key features: (1) high adaptability to dynamic environment by proactive and convex-optimal estimation of nodes' volatile states; (2) minimized task migration conflict probability (upper bound can be limited to 2%) of over-utilized nodes individually shifting surplus loads; (3) a load-status conscious gossip protocol for optimizing distributed resource discovery effect. Based on a real-life user's workload and capacity distribution, the simulation results show that our approach could get significantly improved throughput with 23.6-47.1% reduction on unprocessed workload compared to other methods. We also observe high scalability of our solution under dynamic peer-churning situations.

This paper abstracts the problem of network nodes discovering one another in a network of unknown size using all-to-all gossip. The problem is studied in terms of evolving directed graphs where each vertex represents a participating node and each edge represents one node’s knowledge about another. Ideally, such a graph has diameter one, i.e., each node knows all others. Nodes share their knowledge by sending gossip messages. Gossip among the nodes allows them to discover one another, decreasing the diameter of the graph. Here this problem is considered in several synchronous settings under different assumptions about the ability of the participating nodes to communicate. Specifically, the following aspects of communication are considered: (1) the ability of the nodes to multicast gossip messages, and (2) the size of the messages. The results describe the lower and upper bounds on the number of synchronous rounds required for the participants to discover each other. A particular question of interest is: if the network size is unknown, how does a node know that it has discovered all other nodes? Given a weakly-connected graph describing the initial knowledge of the nodes, every node in our algorithm can stop the discovery process knowing that there are no unknown nodes—this is done without any prior knowledge of the total number of nodes participating in the computation.