The level??’1 mirrors of the root are those that would have
been selected if d had been 1; these level ??’ 1 mirrors are the children of the
root in this hierarchy. A similar recursive definition can be applied to levels
greater than 1. Notice that di?®erent events will have di?®erent hierarchies since
the spatial decomposition is defined for a given event name.
A node that generates an event would, instead of storing the event at the
root (or home) node, store it at the nearest mirror. This mirror is computed
using simple geometric operations, knowing d and the boundary of the sensor
network (the rectangle encompassing the sensor network). Thus, the root node
is no longer a hot-spot since events are distributed across the network. However,
queries for a particular event now have to be directed to all the mirrors.
The mirror hierarchy is used for this in a straightforward way: the query is
sent directly to the root, which then forwards the query to its children, and so
on, until it arrives at all the mirrors. Replies make their way up the hierarchy
in the reverse direction, and may get aggregation along the way.
Note that, structured replication does not replicate data at multiple nodes;
rather, in some sense, the home node for an event is now replicated in several
sub-regions to alleviate hot-spots. This is how the name comes.
root
level 2 (50, 0)
(50, 50) (0, 50)
(0, 0)
mirrior points
level 1
mirror points
Fig.
Pages:
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487