Finally, when the information being sought is obtained, the query
response is routed back to the original querying node. This is illustrated in
Figure 1.
Fig. 1. Illustration of ACQUIRE.
One interesting observation about the ACQUIRE querying mechanism is
that the look-ahead parameter d essentially allows for tuning across a wide
range of behaviors. When d = 0, then the query is simply forwarded along
some path until it is resolved (e.g., a random walk query, or a geographic
389
Active node
Query forwarding
Cache updates
Returned response
LEGEND
Bhaskar Krishnamachari
trajectory based query). When d is large enough to be the diameter of the
network, then the query is essentially performed as a network-wide cachebased
flooding.
A key question in this setting that we shall address with first-order analysis
is what should determine the optimal setting of this look-ahead parameter d. It
turns out that, since caching is employed, this is a?®ected primarily by the ratio
of updates to queries, which we denote by c. (When c = 0.01, for example, on
average one update is requested every 100 queries. Alternately, we could say
that the cache at each active node remains valid on average for 100 queries).
This parameter quantifies the level of dynamics in the environment relative
to the rate at which queries are posed.
We will use, as the metric of interest, the average total number of transmissions
required to resolve a query.
Pages:
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620