Suchergebnisse

Data-intensive applications are one of the largest customers of high throughput computing environments. As clearly shown by the BigData raise, there is a widely recognised urge for processing large amount of data, universallyconsidered the "new oil", from which to extract wisdom, knowledge and, more in general, information of very different kinds, ranging from commercial to financial, from medical to political hints. This extreme intrinsic richness andheterogeneity of data is paired with the extreme complexity in its storing and processing.

Data-intensive applications are one of the largest customers of high throughput computing environments. Asclearly shown by the BigData raise, there is a widely recognised urge for processing large amount of data, universallyconsidered the ?new oil?, from which to extract wisdom, knowledge and, more in general, information of very different kinds, ranging from commercial to financial, from medical to political hints. This extreme intrinsic richness andheterogeneity of data is paired with the extreme complexity in its storing and processing.

The information flow inside a P2P network is highly dependent on the network structure. In order to ease the diffusion of relevant data toward interested peers, many P2P protocols gather similar nodes by putting them in direct contact. With this approach the similarity between nodes is computed in a point- to-point fashion: each peer individually identifies the nodes that share similar interests with it. This leads to the creation of a sort of "private" communities, limited to each peer neighbors list. This "private" knowledge do not allow to identify the features needed to discover and characterize the correlations that collect similar peers in broader groups. In order to let these correlations to emerge, the collective knowledge of peers must be exploited. One common problem to overcome in order to avoid the "private" vision of the network, is related to how distributively determine the representation of a community and how nodes may decide to belong to it. We propose to use a gossip-like approach in order to let peers elect and identify leaders of interest communities. Once leaders are elected, their profiles are used as community representatives. Peers decide to adhere to a community or another by choosing the most similar representative they know about. Index Terms-distributed clustering; peer-to-peer;