270 likes | 393 Views
This paper introduces a novel distributed spatio-temporal indexing framework designed for tracking mobile objects, specifically focusing on plume tracking within sensor networks. It tackles the challenges of query evaluation concerning location, size, and plume distribution, utilizing a hierarchical quad-tree partitioning method to structure the sensor network. The modeling includes adaptive querying through the Smallest Common Ancestor (SCA) algorithm and direct query methods, allowing for dynamic updates based on the plume's movement, ensuring efficient and timely data retrieval in various scenarios.
E N D
Anand Meka and Ambuj Singh UCSB, 2005 DIST: A Distributed Spatio-temporal Index Structure for Sensor Networks
Introduction • Address the problem of plume tracking (in general, tracking of a mobile object) in a sensor network. • Design an analytical model to evaluate the expected cost based on the query location, query size and plume distribution.
Spatio-temporal Indexing • The sensor network is hierarchically decomposed into levels and a quad-tree partitioning (called cells) at each level. • A distributed indexing scheme exploits the plume's locality in space and time using a hierarchical index.
Spatial decomposition of the network • A plume can be mapped to a specific set of cells S at level α that contains it. • α and S can change dynamically as in α(t) and S(t) • α does not change by more than one in two consecutive time instants. • The plume does not skip across the neighbors of a cell between two consecutive time instants.
Q:[42,65] X [42,48] X [t5, t11] • Return F, G
Shape summaries & update propagation • Every leader stores an index or a set of disjoint time intervals over which the plume was inside its cell. Each time interval has a begin and an end time instant such as [t1,t2]. • Assume that a plume's shape is continuously tracked and stored at specific sensor nodes called repository nodes.
Information maintaining • At each time instant t, a repository node senses the plume and computes α(t) and S(t). • How can the repository node know the α and S ?? • A repository node sends a message (id,t) to the leader of each cell c in S(t). • l(c) updates information. • Any neighboring cell d of c that had an open index at time t-1 ends its most recent time interval by inserting t-1. • Who notify those cells ??
Range Query Algorithm - SCA • Smallest Common Ancestor algorithm • The query originator determine the spatial cells at the level ε that are intersected by the query. • Determines the smallest common ancestor sca of these cells. • Transmits the query to the sca using an GPSR.
Direct query algorithm • Query originator decomposes the query's spatial extent into cells at level ε, and directly queries these cells and all their ancestors. • Constructing a spanning tree (ST) at each level. • The query originator constructs a communication graph and finds a ST.
Adaptive querying • Both the SCA and Direct query algorithms have their advantages and disadvantages. • SCA is effective in the case of a query with a large spatial range. • Direct query – small spatial extent • Adopting the better of the two schemes depending on the query location, query size, and plume distribution on a per-query basis.
Performance Evaluation • Simulation and mobility models • Cloud model: the centre of mass of the plume performs a random walk. • Gaussian plume dispersion model: the concentration of the plume perpendicular to the direction of the wind velocity follows a Gaussian distribution.
Performance Evaluation • Update costs
Performance Evaluation • Query costs
Conclusion • Direct query • SCA query • Adaptive scheme