For example, N sensors are deployed in an area to
measure temperature. To compute the average temperature accurately, the
data from all the N sensors should be obtained and aggregated, which will
cost a lot of power. Random sampling techniques can be adopted in this case.
By satisfying user-requested-accuracy, only the data fromM sensors (M < N)
should be used. Thus, the cost is reduced dramatically. A traditional database
does not support such a technique.
The sixth di?®erence is that, query processing techniques in a traditional
database system are not suitable for sensor networks. The reasons are as
follows:
??? The query optimization technique in a traditional data-base system is
based on a fixed cost model and statistical information. There is no reliable
statistical information about data in sensor networks, and it is very di?±cult
to forecast the action of data streams from sensors. The query plan in
sensor networks must minimize the power consumption and be suitable
for the actions of data streams.
??? A traditional database system locks the data and unlocks them when errors
occur. This is unsuitable for sensor networks since the actions of data
streams are uncertain and the query may last for a long time.
??? A traditional database system operates on an existing and fixed database.
Answers for queries are fixed. In sensor networks, the objects operated on
are infinite and uncertain data streams.
Pages:
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467