3.

MINIMUM DISTANCE PACKET FORWARDING

3.1 FRAMEWORK OVERVIEW
The idea behind MDPF is to use routing table information for visiting
nodes in the order of shortest distance (hop counts). As implied, this
requires valid routing information, hich could be handled through a
proactive routing protocol such as the D!D" protocol or an on#demand
reactive routing protocol, li$e A%D". Assuming that the set of nodes that
hold the search information is $non to all nodes in the ireless netor$,
and refer to these nodes as the search nodes, or !&s. 'e emphasi(e though
that a requesting node hich is interested in a particular data item does not
usuall) $no hich specific !& holds the location of the data item, and
therefore, it must search for it in the !&s.
3.2 BASIC OPERATION
According to MDPF, the client uses the information in the routing
tables to send its request to the nearest !&. *f an !& does not have the
requested data, it also uses MDPF and forards the request to the nearest
unvisited !&. Fig. + shos to e,ample scenarios. &odes request database
data, hich ma) be cached in an) of the caching nodes (-&s). The search
nodes (!&s) cache previousl) submitted requests (queries), and for each
such quer), an !& maintains a reference to the result that resides on a -&.
*n the first scenario, the client submits its request to the nearest !& (!&.),
hich does not have a matching quer). The request is then forarded in
accordance ith MDPF through !&+ and !&/ before it arrives to !&0, here
a match is found. 1sing the reference that is stored along ith the cached
quer), the request of the client is forarded to the -& that stores the result.
This -& sends the result to the client hose address is found in the
forarded pac$et. *n the second scenario, no match is found in the !&s, and
so, the last visited !& (!&2) forards the request to the data server via the
access point. The server retrieves the result and sends it directl) to the
client, hich, in turn, as$s !&. (being its nearest !&) to cache the quer). *t
is noted that the node at hich the client requested the data item that as
retrieved from the outside data server becomes a -& for this particular item.
Fig. 1. To scenarios for request forarding3 !cenario + corresponds to a
hit and includes steps +#/, 2a, and 4a. !cenario 0 describes a miss and
includes steps +#/, /5, /55, 2b, 4b, and 6b.
'hen a proactive routing protocol is emplo)ed, MDPF can readil) use
the routing information to choose the !& that requires the minimum number
of hops from the set of unchec$ed !&s. 7oever, hen an on#demand
routing protocol, such as the A%D" or D)namic !ource 8outing (D!8)
protocols, is in place, the routing information to the nearest unvisited !&
must be discovered on demand if necessar) (i.e., if it is not cached, or if it
is cached but not fresh) and $ept in the routing table for a certain period of
time before it e,pires. More specificall), hen a reactive routing protocol is
emplo)ed, MDPF or$s as follos. 9ach node e,amines its routing table to
find if the routing information to the unchec$ed !&s is present and valid. *f
)es, the node acts as in the proactive case and chooses the !& ith the
minimum number of hops to reach. *f the node finds that its routing table
does not contain the routing information for one or more unchec$ed !&s, it
broadcasts an !& Discover) Pac$et (!&DP) containing the list of unchec$ed
!&s and a sequence number to all its neighbors.
'hen a neighbor receives an !&DP the first time, it chec$s its routing
table for the presence of one or more unchec$ed !&s. *f it $nos of such
!&s, the neighbor sends the routing information of these !&s to the
requesting node. 9lse, the neighbor broadcasts (forards) the !&DP to its
on neighbors. *n order to prevent the possibilit) of flooding the netor$
ith pac$ets, the !&DP contains a hop limit $ that denotes the ma,imum
number of hops aa) from the source that the !&DP can be sent to. The
value of $ depends on the netor$ si(e, the total number of nodes, the
transmission range, and the number of current unchec$ed !&s. For e,ample,
for a +:;;; < +:;;; m0 netor$ containing +;; nodes, and hen the
number of unchec$ed !&s is 6, the netor$ diameter is appro,imatel) +/
hops and $ could be set to +/=6 > 0 hops (assuming that the !&s are
uniforml) spread throughout the netor$). As the number of unchec$ed !&s
decreases, $ increases, and vice versa. 'hen this number is +, $ ill be
equal to the netor$ diameter.
Finall), the !&DP source node aits for time < (e.g., ;.+ sec),
e,amines the routing information to the !&s it received, then chooses the
!& ith minimum number of hops to reach, and forards the search pac$et
to it. *t also adds the routing information to its routing table for future use.
3.3 Evaluati! M"t#$lg%
The ob?ective of this paper is to propose a message forarding
algorithm for search applications and anal)(e its performance. 'e focus on
the anal)sis of the hop count to reach the !& having the desired data, and
to traverse all the !&s. 'e also consider an important metric that concerns
fairness, namel), the average traffic load e,perienced b) the different !&s.
*n addition to the e,perimental evaluation using the ns0 simulation softare,
e anal)(e the s)stem5s performance using anal)tical derivations in the case
of traffic load and numerical anal)sis in the case of hop count. The reason
for this is that simulation b) itself does not ala)s )ield completel) reliable
results.
3.& R"'ult' R"lia(ilit%
!imulation approaches usuall) suffer from a lac$ of reliabilit) because
it is difficult to prove that the samples ta$en out of a certain probabilit)
distribution are indeed t)pical, or that the sampling distribution of their
mean closel) follos a @aussian la. For e,ample, probabilit) distributions
ith a high $urtosis ma) have sample means hich are not close to the
actual mean of the distribution. All these problems and others affect the
reliabilit) of simulation in general, hile the anal)tical solution is usuall)
reliable. !ince the MDPF algorithm (or a ver) close variant thereof) has, in
fact, alread) been studied in -omputer !cience under the name of A&earest
&eighborB heuristic for the traveling salesperson problem, several papers can
be found in the literature on the sub?ect.
1nfortunatel), the problems in this area turned out to be so difficult
that researchers ho attempted to tac$le similar problems anal)ticall) did so
under unrealistic simplif)ing assumptions such as considering all distances
beteen pair of points statisticall) independent from each other (i.e., even
ignoring the triangular inequalit)), hile some other researchers obtained
anal)tical solutions on much simpler problems, for e,ample, b) restricting
themselves to the one#dimensional case, leaving the to#dimensional
problem unsolved due to its difficult). *n contrast, the probabilistic results
for similar problems, hich are considered the most reliable, have been
obtained through simulation. !till, hoever, e do not give up on the
anal)tical solution of the nearest !& search problem. *n this regard, it might
be useful to point out the main challenge that ma$es the problem a difficult
one, even under the infinite node densit) assumption.
First, it is not hard to obtain an e,pression for the probabilit)
distribution of the closest !& to a given random !&. 'e assume that the
!&s are randoml), uniforml), and independentl) distributed on the
considered area, and therefore, the probabilit) distribution function of the
distance to the closest !& is simpl) the same as the distance sample
minimum. The sample minimum has a closed#form formula hich could be
applied. Cut the difficulties start to appear hen e ish to determine the
probabilit) distribution of the distance beteen the second and the third !&.
The main problem here is that the distribution of available !&s around the
second !& is not independent from the position of the first !&. *ndeed, since
the second !& as the closest one to the first, it means that the second !&
is on the boundar) of a dis$ centered at the first !& hich is empt) of an)
!&. As one reaches the third, fourth, and nth !&, the empt) area becomes
an ever more complicated union of dis$s, ma$ing it difficult to obtain a
provabl) accurate anal)sis.