Figure 15 illustrates a pipelined
aggregate in a small sensor network.
The most significant drawback of a pipelined aggregate is that a number
of extra messages are transmitted in order to get the first aggregation of all
the sensors. Figure 15 is an example. It requires 22 messages for the first
result, since every aggregating node needs to transmit a message in each time
interval. The non-pipelined approach only needs to propagate 10 messages, five
down and five up along the network. In this example, after 12 initial messages
overhead, it only needs five messages to produce each additional aggregate at
the rate of one update per time interval. The pipelined aggregation can still be
optimized to reduce the transmitting overhead. One option is that each node
does not transmit any data until the aggregate of its subtree is changed, and
intermediate nodes assume aggregates of their child nodes remain unchanged
unless they receive di?®erent values. This method e?±ciently reduces the amount
of messages sent by nodes.
Channel sharing during aggregation [12]
While discussing pipelined aggregate, the fact that sensor nodes use a shared
channel to communicate with each other is ignored. Messages are broadcasted
and all sensor nodes within the range can receive them. Taking the advantage
of a shared channel, some techniques can reduce the tra?±c and increase the
accuracy of aggregation in the face of transmission failures.
Pages:
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506