%0 Conference Paper
%B The 11th International Workshop on. Large-Scale and Distributed Systems - LSDS-IR
%D 2007
%T Taming hot-spots in dht inverted indexes
%A Nuno Lopes
%A Carlos Baquero Moreno
%C Amsterdam, The Netherlands
%I Citeseer
%V 7
%X <p>DHT systems are structured overlay networks capable of using<br /> P2P resources as a scalable platform for very large data storage<br /> applications. However, their efficiency expects a level of uni-<br /> formity in the association of data to index keys that is often<br /> not present in inverted indexes. Index data tends to follow non-<br /> uniform distributions, often power law distributions, creating in-<br /> tense local storage hotspots and network bottlenecks on specific<br /> hosts. Current techniques like caching cannot, alone, cope with<br /> this issue.<br /> We propose a new distributed data structure based on a decen-<br /> tralized balanced tree to balance storage data and network load<br /> more uniformly across all hosts. The approach is stackable with<br /> standard DHTs and ensures that the DHT storage subsystem re-<br /> ceives an uniform load by assigning fixed sized, or low variance,<br /> blocks.</p>
%8 February
%> https://haslab.uminho.pt/sites/default/files/cbm/files/acmlsdsir07.pdf