MARQUES: Distributed multi-attribute range query solution using space filling curve on DTHs

NSysS(2015)

引用 3|浏览12
暂无评分
摘要
This paper proposes a distributed peer-to-peer data lookup technique on DHTs in order to serve range queries over multiple attributes. The scheme, MARQUES, uses space filling curves to map multi-attribute data points to a one-dimensional key space and thus effectively converts multi-attribute range queries into a consecutive series of one-dimensional keys. These keys are then used to place or lookup data objects over a DHT. Space filling curves preserve locality of attribute values and thus helps greatly in facilitating range queries in terms of the number of nodes to be searched to serve a given range query. MARQUES, although can be instrumented with any space filling curve, has been implemented with two curves, namely Z-order curve and Hilbert curve, and uses a multi-level variant of Chord, a popular DHT, as its underlying overlay. Simulation results on OMNET++ show that MARQUES successfully answers range queries with significant efficiency in terms of message overhead and query latency.
更多
查看译文
关键词
distributed processing,file organisation,query processing,dth,hilbert curve,marques,omnet++,z-order curve,attribute value locality preservation,data objects,distributed hash tables,distributed multiattribute range query,distributed peer-to-peer data lookup technique,message overhead,multiattribute data point mapping,multilevel chord variant,one-dimensional key space,overlay,query latency,space filling curve,indexes,computer science,routing,protocols,distributed databases
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络
Chat Paper
正在生成论文摘要