However, when node C sends back a response
message, node P cannot receive the message since it may be lost during transmission.
As a result, node P assumes it has no child node and only returns its
own data. Consequently, the entire subtree below node P is excluded from the
computation and then probably a wrong aggregate result may be produced.
In fact, any subtree of the network may fail in this way, i.e., it is impossible
to guarantee the accuracy of the final result.
One solution to the above problem is to validate aggregations by computing
them multiple times. The simplest method is to perform aggregation
at the root node of the network multiple times. Users can observe those aggregates
and make a reasonable decision about which is the true value. The
problem of this technique is that it requires re-sending the aggregate request
along the network multiple times, causing a large communication cost and a
large energy consumption cost. Users must wait for an aggregation interval
for each additional result, which prolongs the query response time.
Pipelined aggregation is an e?®ective technique which is used to overcome
the problem of accuracy of the aggregation in sensor networks. It divides time
into many intervals. Assume the length of an interval is i seconds. During
each interval, the sensor that has received the aggregate combines its local
readings with the values its child nodes reported during the previous interval,
and then propagates the partial aggregation to its upper level in the network.
Pages:
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504