Review: Epidemic-style Management of Semantic Overlays for Content-Based Searching
Spyros Voulgaris and Maarten van Steen, "Epidemic-style Management of Semantic Overlays for Content-Based Searching," EuroPar 2005, Aug 2005
According to the authors, this is one of the pioneer works in creating semantic overlays using epidemic protocols. Semantic overlays are application level interconnection of nodes where semantically closed neighbors are connected with each other. The semantic proximity depends on the application of the network. For example, in case of a file sharing application, the amount of common files two nodes share can determine this measure (This is the metric used in this paper).
The paper clearly analyses the architecture of an epidemic protocol-based semantic network. Compared to the earlier proposals (according to the authors) which assumes stable network, the epidemic protocol-based architecture performs very well in environments under frequent churning. In this type of semantic overlay, each node maintains a set of nodes, called its semantic view, out of which it randomly selects a peer periodically and exchange its semantic information. Transitivity can be reasonably assumed among semantic neighbors, that is, if A is semantically closer to B that is semantically closer to C, A can be expected to be semantically closer to C. This notion helps in refining a node's semantic view by gossiping with its semantic neighbors periodically.
However, this leads to semantic clustering which prevents a node to discover nodes in another semantic cluster that become semantically closer recently. Therefore, this paper proposes a two-tier gossip overlay for constructing the semantic overlay where the top, actual semantic overlay depends on the bottom random-peer sampling overlay who can propose random nodes in the system to communicate with. The random-sampling overlay is based on epidemic-protocol as well.
The paper shows through via simulations that the convergence of the system into semantic clusters is fast and it can adjust itself with changes in the semantics due to node failures or changes in the file cache.
According to the authors, this is one of the pioneer works in creating semantic overlays using epidemic protocols. Semantic overlays are application level interconnection of nodes where semantically closed neighbors are connected with each other. The semantic proximity depends on the application of the network. For example, in case of a file sharing application, the amount of common files two nodes share can determine this measure (This is the metric used in this paper).
The paper clearly analyses the architecture of an epidemic protocol-based semantic network. Compared to the earlier proposals (according to the authors) which assumes stable network, the epidemic protocol-based architecture performs very well in environments under frequent churning. In this type of semantic overlay, each node maintains a set of nodes, called its semantic view, out of which it randomly selects a peer periodically and exchange its semantic information. Transitivity can be reasonably assumed among semantic neighbors, that is, if A is semantically closer to B that is semantically closer to C, A can be expected to be semantically closer to C. This notion helps in refining a node's semantic view by gossiping with its semantic neighbors periodically.
However, this leads to semantic clustering which prevents a node to discover nodes in another semantic cluster that become semantically closer recently. Therefore, this paper proposes a two-tier gossip overlay for constructing the semantic overlay where the top, actual semantic overlay depends on the bottom random-peer sampling overlay who can propose random nodes in the system to communicate with. The random-sampling overlay is based on epidemic-protocol as well.
The paper shows through via simulations that the convergence of the system into semantic clusters is fast and it can adjust itself with changes in the semantics due to node failures or changes in the file cache.
0 Comments:
Post a Comment
<< Home