Review: The Case For a Hybrid P2P Search Infrastructure
Boon Thau Loo, Ryan Huebsch, Ion Stoica and Joseph M. Hellerstein, "The Case for a Hybrid P2P Search Infrastructure," IPTPS 2004, pp 141-150
The paper is of two parts: in the first part, it analyses the performance of gnutella network in discovering rare items; and in the second part, it proposes a hybrid overlay to address the problem in gnutella network (or generally in an unstructured network).
The paper probes the gnutella network using mutliple vantage points and shows that the gnutella network performs badly in discovering rare items. It points out that structured networks are the exact opposite in nature, performing better with rare items, but introducing too much overhead with common items. This justify a hybrid architecture combining both the unstructured and structured overlays. In the hybrid architecture, the queries for common items are propagated in the unstructured overlay while the structured overlay is used for propagating queries for rare items.
The problem in this hybrid overlay is to identify which items are rare and which are not. The proposed architecture addresses the problem by connecting only the ultrapeer (similar to the superpeer in KaZaA) nodes in gnutella to join the structured overlay. These nodes can observe the results for the queries generated by the leaf nodes and decide which items are rare and which are not. Leaf nodes uses the unstructured overlay for queries and, in case of no result within a certain time, it relaunches the query on the structured overlay. The paper provides figures to prove that the search performance is improved in the hybrid.
Even though the paper talks about a hybrid of unstructured and structured networks, it consider a gnutella type of unstructured network which, with the use of ultrapeer, not a pure unstructured network. On the other hand, it can be argued that considering just the ultrapeers it is an unstructured network.
The paper is of two parts: in the first part, it analyses the performance of gnutella network in discovering rare items; and in the second part, it proposes a hybrid overlay to address the problem in gnutella network (or generally in an unstructured network).
The paper probes the gnutella network using mutliple vantage points and shows that the gnutella network performs badly in discovering rare items. It points out that structured networks are the exact opposite in nature, performing better with rare items, but introducing too much overhead with common items. This justify a hybrid architecture combining both the unstructured and structured overlays. In the hybrid architecture, the queries for common items are propagated in the unstructured overlay while the structured overlay is used for propagating queries for rare items.
The problem in this hybrid overlay is to identify which items are rare and which are not. The proposed architecture addresses the problem by connecting only the ultrapeer (similar to the superpeer in KaZaA) nodes in gnutella to join the structured overlay. These nodes can observe the results for the queries generated by the leaf nodes and decide which items are rare and which are not. Leaf nodes uses the unstructured overlay for queries and, in case of no result within a certain time, it relaunches the query on the structured overlay. The paper provides figures to prove that the search performance is improved in the hybrid.
Even though the paper talks about a hybrid of unstructured and structured networks, it consider a gnutella type of unstructured network which, with the use of ultrapeer, not a pure unstructured network. On the other hand, it can be argued that considering just the ultrapeers it is an unstructured network.
posted by Unknown at
9:35 AM
0 Comments:
Post a Comment
<< Home