Wireless Sensor Networks : Technology, Protocols, and Applications

PERFORMANCE OF TRANSPORT CONTROL PROTOCOLS 243
20
15
Ratio
10
High.noCODA.E.Tax
Low.noCODA.E.Tax
5
High.CODA.E.Tax
Low.CODA.E.Tax
0
Network Size
(# of nodes)
10 30 50 70 90 110
Figure 7.1 Energy tax in CODA as a function of network size for high- and low-data-rate
traffic. The difference between the data points with and without CODA indicates the energy
saving achieved by CODA. (Based on data from [7.1].)
between two neighboring nodes. In the noncache case, however, retransmissions
may occur in h hops, and therefore more total energy is required. The cache
point is defined as the node that copies transmitted packets locally for a certain
time period; and the loss point is defined as the node at which packets are
dropped due to congestion. Let’s define the retransmission path length ðl p Þ as
the number of hops from the caching node to the node where loss occurs. Therefore,
in the noncache case, l p ¼ h 1 ,whereh 1 is the number of hops from the loss
point to the source node. In the cache case, l p can be 1 if lost packets are found
in neighboring nodes. Because sensor nodes have limited buffer space, packet
copies can be stored for only a limited time period. Therefore, l p in the cache
case may be larger than 1 but still smaller than h 1 ð1 < l p < h 1 Þ. In cachebased
recovery, different algorithms may have a different retransmission path
length l p and introduce different energy efficiency.
In cache-based recovery, each packet is stored at every intermediate node that it
visits until its neighboring node receives the packet successfully, or when a timeout
occurs (whichever is sooner). In this case it is likely that l p is very close to 1. Another
option is to distribute caching so that packet copies are scattered among intermediated
nodes. Each packet is stored in only one or several intermediate nodes.
Distributed caching might have a longer l p than regular caching (but still smaller
than in the noncache case) and requires less buffer space than regular caching.
RMST [7.3] investigated the performance of various loss recovery mechanisms
that may provide reliability through the link, transport, and application layers.
Figure 7.2 from [7.3] compares the performance of hop-by-hop and end-to-end
loss recovery in the transport layer. The comparison is made in terms of the number
of transmissions required to send 10 packets across a network in 10 hops. As shown