Image Processing Reference
In-Depth Information
Directed Diffusion does not directly decrease the duty cycle of the nodes (i.e., nodes do not go to
sleep), so energy efficiency is only achieved by reducing the number of packets transmitted and per-
formingdataaggregation.histechniqueiseicientwhentheidlepowerconsumptionismuchlower
than that required for transmitting and receiving. However, in many real cases of devices character-
ized by high power consumption in idle states (i.e., with CSMA protocols) Directed Diffusion does
not achieve high energy savings. Despite this, with respect to the SPIN protocol family, Directed
Diffusion features lower overheads in terms of packets sent, as SPIN needs explicit negotiation
for each packet, while Directed Diffusion uses periodic interests and gradients perform implicit
negotiation. Lower overheads also imply higher energy efficiency. However, in Directed Diffusion
the number of data packets is higher than in SPIN, due to the existence of multiple paths, so the
energy consumption of Directed Diffusion also depends on the size of data packets as compared to
negotiation packets.
In [Gad] the Directed Difusion approach was compared with classical MANET routing proto-
cols,suchastheAd-hocOn-demandDistanceVectorprotocol(AODV)[Perk]andtheOptimized
Link State Routing protocol (OLSR) [Jacq], and with another data-centric protocol, Two-Tier Data
Dissemination (TTDD) [Ye], for both performance and energy efficiency. Simulation results show
that AODV regularly outperforms others in terms of packet delivery ratio, latency, and energy effi-
ciency. The lower efficiency of the Directed Diffusion protocol is probably due to the higher data
redundancy. In fact, although there is only one reinforced path, multiple low data rate links are main-
tained. In addition, the scenarios tested only comprise a small number of nodes producing data, so it
is more suitable for classical ad-hoc networks than for WSNs. On the other hand, WSNs may comprise
a very large number of sensing nodes. In such scenarios, maintaining a routing table with an entry
for each data flow may be too expensive for resource-constrained nodes such as typical WNS nodes,
so the scalability of localized
∗
WSN protocols (such as data-centric or location-based approaches) is
required [Estr]. Furthermore, WSN nodes are generally more faulty than other networks, as the
network may operate in harsh environmental conditions. In such conditions, the Directed Difusion
approach can provide higher fault tolerance as multiple paths are maintained for each data flow.
The Directed Diffusion approach can definitely be useful for query-based sensor networks, as
semantic information is used for forwarding. The drawback of this approach is that the naming
scheme is strongly application-specific. However, a WSN is not usually a general-purpose network,
so this is not a main concern.
7.3.3 Rumor Routing Protocol
Rumor Routing [Bra] is a probabilistic data-centric routing protocol for large-scale WSNs contain-
ing several thousands of nodes, which is inspired by the Directed Diffusion approach. he Directed
Diffusion approach disseminates queries through networks by means of (controlled) looding. his is
eicientiftheamountofdatatobetransmittedisnotsmalloriftherearealargenumberofevents.In
some cases, i.e., when there are few events and many queries, event broadcast may be more efficient
than query broadcast. Rumor Routing is something in between event broadcast and query broadcast,
soitcanbeusefulinthemiddleoftheregiondepictedinFigure..
In Rumor Routing, each node maintains a neighbor table as well as an event table. When a node
noticesanevent,itprobabilisticallygeneratesanagent.Anagentisalong-livedpacketthattravels
through the network in order to propagate information about the sensed events to distant nodes.
When a node receives such a packet, it updates the event table with the distance and the forwarding
∗
Localized protocols maintain and use only information about neighbor nodes; no overall network or transmission flow
knowledge is required.
