Consider
a sensor network where each sensor node periodically generates temperature
samples. Viewed across the entire network, the collection of temperature
samples can be represented by a 2D array of temperature values indexed by
location and time. If all these values could be collected in a central location,
then it is clear that one could compute wavelet coe?±cients on the data to
e?±ciently answer spatial-temporal queries described above. However, the design
of the DIMENSIONS system makes the observation that these wavelet
coe?±cients can be computed and stored far more e?±ciently in a distributed
fashion. This capability rests on the two properties: resolution in the spatial
dimension corresponds to di?®erently sized regions of the sensor field, and that
lower resolution wavelet coe?±cients can be computed from higher-resolution
coe?±cients. DIMENSIONS constructs a storage hierarchy using data-centric
storage concepts, and stores successively lower resolution coe?±cients in the
higher levels of the hierarchy. Given a number of levels of resolution d, the
system e?®ectively divides the geographic region occupied by the sensors into
d levels recursively. The 0-level has only one region, that is, the area covered
by the sensor network. The i-level has 2i sub-regions. The d-level has
2d sub-regions. The size of each sub-region in i-level is four times the size of
a sub-region in i + 1-level. To construct the 1-level from the 0-level in the
hierarchical structure, the system first hashes the names of the datasets corresponding
to 0-level to a location in sub-region of 0-level.
Pages:
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490