<<

. 23
( 87 .)



>>

Current ad hoc network architectures do not take into account this difference, and im-
plement in ad hoc stations the same functionalities of a router. Specifically, packets re-
ceived from the wireless medium are delivered to the IP layer where a route lookup is per-
formed based on the destination IP address (steps 1“3 in the left-hand part of Figure
3.29). If the packet is not destined to the station itself, it is passed down to the network in-
terface to be retransmitted (steps 4 and 5 in the left-hand part of Figure 3.29).
The difference, from the forwarding standpoint, between an ad hoc station and a router
has been recently pointed out by Acharya et al. [1], who have also proposed an architec-
ture for efficient packet forwarding at stations in multihop ad hoc networks.
The first question addressed in [1] is which is the best architecture for forwarding a
packet in ad hoc networks. The answer is highlighted in Figure 3.29. The left-hand side of
the figure depicts the legacy approach for forwarding packets; the right-hand side shows
the new approach proposed by the authors. In the latter case, the forwarding is completely
managed by the network interface card (NIC). Upon receiving a packet, the NIC (by ex-
ploiting some local information) determines whether or not the packet has to be retrans-
mitted. Only packets destined to the station itself are passed to higher-level protocols. Due
to its behavior, the proposed architecture has been named cut-through architecture [1].
The cut-through architecture provides several advantages that can be classified into
two categories. The first category includes advantages that are not related to a specific
MAC protocol, where advantages belonging to the second category are strongly related to
the random access scheme and the RTS-CTS mechanism used in the IEEE 802.11 MAC
protocol. The following advantages belong to the first category:

1. The NIC does not need to interrupt the CPU for packet processing. This could lead
to a considerable power saving if, for example, the station is used only for packet
forwarding purposes. The CPU needs to wake up only for processing route updates.
2. Delays for transferring data from the NIC to the host, and vice versa, are avoided.




A B

R A
R B




Figure 3.28. Packet forwarding in a router in a wired network (left) and in an ad hoc network (right).
109
3.6 EVOLUTION OF IEEE 802.11b FOR AD HOC NETWORKS




Host Host
3 IP
Layer

4
2

NIC 2
NIC
5
1 1 3


Figure 3.29. Forwarding in ad hoc stations: legacy approach (left) and NIC-based forwarding
(right).



3. Local traffic does not further delay forwarding traffic. In legacy IEEE802.11 archi-
tecture, at the forwarding station, the packet is transferred to the main memory by
the NIC. The host CPU is notified (e.g., via interrupts) for further processing of the
packet by the IP protocol stack running on the host CPU. The host software (IP pro-
tocol stack) would typically queue up the packet in a transmission queue (together
with the locally generated packets) and select packets for transmission based on a
scheduling algorithm (typically, FIFO). Thus, packets generated by applications
running at the station can overtake packets to be forwarded. This produces an in-
crease in the end-to-end delay.

As the other advantages are strictly related to the IEEE 802.11 MAC protocol, they can
be better understood by first considering the operations performed at the MAC layer by
stations A, B, and R in the scenario depicted in the right side of Figure 3.28.

3.6.1 Forwarding Operation: Legacy approach versus
NIC-Based Approach
Let us consider the case shown in Figure 3.30, where A is the upstream station, R is the
forwarding station, and B the downstream station. In the legacy approach, the data deliv-
ery from A to B involves two separate and independent transmissions. For each transmis-
sion, the IEEE 802.11 MAC protocol (including the RTS/CTS mechanism) is used.
Specifically, with reference to Figure 3.30, the transmission from A to R is first per-
formed. At R, the packet is passed to the IP protocol, processed by the IP software, and
passed down to the NIC for transmission to B. At this time, R has to repeat the same proce-
dure executed by A during the transmission to R. Note that the two transmissions (from A
to R and from R to B) are independent of each other from the channel access standpoint.
It is worth noting that after the first RTS/CTS exchange, stations A and R has control
of the channel and no other station in the transmission range of A and/or R can access the
channel. However, this control is immediately lost after the ACK transmission from R to
A. Clearly, from the Station R standpoint, it is not very wise to release the channel control
and immediately after compete again for gaining control of the channel. It would be better
for R to maintain in exclusive control of the channel. In this case, the transmission to B
would be done without contention, thus improving the bandwidth utilization (there would
110
A to R transmission
Contention
DIFS SIFS
R to B transmission
RTS DATA
A
Contention
SIFS SIFS DIFS SIFS

CTS ACK RTS DATA
R
SIFS SIFS




Node Processing
CTS ACK
B
NAV RTS NAV RTS
NAV CTS NAV CTS
NAV DATA NAV DATA
A and R can access the channel R and B can access the channel
The other stations must defer the medium access The other stations must defer the medium access

Figure 3.30. Forwarding operations in the legacy approach.
111
3.6 EVOLUTION OF IEEE 802.11b FOR AD HOC NETWORKS


be no bandwidth wastage due to collisions and backoff periods) and minimizing the for-
warding delay. Obviously, Station R should execute the packet forwarding very quickly so
that the transmission from R to B could start immediately after the ACK transmission
from R to A. This can be achieved only if the forwarding operation is performed com-
pletely inside the NIC (right-hand side of Figure 3.29).
Figure 3.31 shows an extension of the IEEE 802.11 MAC protocol to manage packet
forwarding in a more efficient way [1]. The basic idea is to give the highest priority to the
traffic to be forwarded by extending the channel reservation scheme to allow on-the-fly
transmissions. Specifically, upon receiving a frame from A, Station R not only sends back
an ACK frame to A but, at the same time, transmits an RTS frame to further extend the
channel reservation.12 Since the RTS frame transmission occurs while all the other sta-
tions within R™s transmission range are still blocked (due to the previous RTS/CTS ex-
change), station R can immediately get the channel. The extended MAC protocol has been
named Data-driven Cut-through Multiple Access (DCMA).

3.6.2 DCMA MAC Protocol
The DCMA MAC protocol is an extension of the IEEE 802.11 DCF and, as such, it fol-
lows the associated four-way handshake involving the exchange of RTS/CTS/DATA/ACK
frames. As shown in the previous section, the DCMA attempts to replace the two distinct
channel accesses (upstream and downstream) with a combined access. Specifically,
DCMA combines the ACK (to the upstream station) with the RTS (to the downstream sta-
tion) in a single ACK/RTS packet that is sent to the MAC broadcast destination address.
The cut-through approach, proposed in DCMA, fails when the downstream station
(e.g., B in our example) cannot reply to the ACK/RTS (with a positive CTS). In such a
case, the forwarding station then simply queues the packet in the NIC queue, and resumes
the normal IEEE 802.11 channel access method using the exponential backoff to regulate
subsequent access to the shared channel. The channel-contention resolution of DCMA is
same as that of 802.11, with a station remaining silent as long as any of its one-hop neigh-
bors is either receiving or transmitting a data packet. Accordingly, this protocol does not
suffer from any additional penalties, over and above those present in 802.11.
Since DCMA has no notion of future reservations (all access attempts are for immedi-
ate transfer of DATA frames), it does not require any modifications or enhancements to
the 802.11 NAV; a station simply stays quiet as long as it is aware of (contiguous) activity
involving one or more of its neighbours. Any station that overhears an ACK/RTS not ad-
dressed to it merely increments the NAV by the time interval included in the ACK/RTS
message.
In [1] a simulation analysis of the DCMA scheme is also presented. This analysis was
carried out by implementing the DCMA access protocol in the ns-2 simulation tool. Con-
sequently, the ns-2 typical values were used: bit rate of 2 Mbps, transmission range equal
to 250 m, and interfering range equal to 550 m. All transmissions, regardless of the frame
size, were preceded by an RTS/CTS exchange.
In the simulation study, high-rate sources were used to guarantee a never-empty queue
at the transmitter. To avoid the interference of TCP mechanisms, the UDP protocol was
used at the transport layer. The statistics were estimated by considering only the packets
correctly received at the receiver. The forwarding station™s routing tables were preconfig-

12
More precisely, the RTS frame is piggybacked to the ACK frame.
112
A to R transmission
Contention
DIFS SIFS
R to B transmission
RTS DATA
A
SIFS SIFS SIFS

CTS ACK RTS DATA
R
SIFS SIFS

CTS ACK
B
NAV RTS NAV RTS
NAV CTS NAV CTS
NAV DATA NAV DATA
A and R can access the channel R and B can access the channel
The other stations must defer the medium access

Figure 3.31. Forwarding operations in the NIC-based approach.
113
3.6 EVOLUTION OF IEEE 802.11b FOR AD HOC NETWORKS


Table 3.9. Throughput Comparison (in Kbps)
Packet size (bytes)
256 512 1024 1536
802.11 159 200 231 258
DCMA 197 242 282 301



Table 3.10. End-to-End Delay Comparison (in sec)
Packet size (bytes)
256 512 1024 1536
802.11 1.00 1.67 2.59 2.83
DCMA 0.50 0.81 1.31 1.73



ured with the shortest-path routes to their respective destinations. The contents of these
routing tables will be briefly discussed in the following paragraphs. A complete descrip-
tion can be found in [1].
Several configurations are considered in [1]. For the sake of brevity, only one set of
them are discussed here. They refer to a string or chain topology (see Figure 3.8) in which
the distance between successive stations is 250 m. A single flow of UDP packets is trans-
mitted from the leftmost to the rightmost station. Several experiments were conducted by
varying the size of the payload from 256 to 1536 bytes.
The results obtained by considering a 7-hop chain are summarized in Table 3.9 and
Table 3.10. It clearly appears that DCMA improves the throughput by around 20% with
respect to the standard protocol, whereas the delay improvement is more significant, rang-
ing from 63% (1536 bytes) to 100% (256 byte packets).
In [1], the comparison was further extended by considering an increasing number of
hops in the chain. The results obtained are consistent with results presented in Table 3.9
and Table 3.10: the delay reductions with DCMA are significant (on the order of 50%),
whereas throughput improvements are marginal.
Finally, the influence of the offered load was considered. The results obtained are sum-
marized in Table 3.11 and Table 3.12. Specifically, these results are related to a 12-hop
chain and have been obtained by increasing the sending rate at the source from 250 Kbps
to about 500 Kbps. It clearly appears that there are different saturation points for the two
protocols. The IEEE 802.11 MAC protocol has the maximum throughput at around 0.375
Mbps; after this offered load level, the queues start to build up, the end-to-end delay


Table 3.11. Throughput (in Kbps) as a Function of the Offered Load
Offered Load (Kbps)
250 275 300 325 350 375 400 425 450 475 500
802.11 242 267 292 316 340 364 259 273 256 241 235
DCMA 242 267 292 316 340 364 389 413 376 376 374
114 IEEE 802.11 AD HOC NETWORKS: PROTOCOLS, PERFORMANCE, AND OPEN ISSUES


Table 3.12. End-to-End Delay (in sec) as a Function of the Offered Load
Offered Load (Kbps)
250 275 300 325 350 375 400 425 450 475 500
802.11 0.075 0.075 0.075 0.075 0.075 0.075 0.075 0.075 0.75 0.95 1.09
DCMA 0.080 0.080 0.080 0.080 0.080 0.080 1.37 1.53 2.06 2.02 2.08




shows a significant increase, and the throughput decreases. On the other hand, DCMA has
the maximum throughput at around 0.425 Mbps. Furthermore, after the saturation point
DCMA shows a more stable behavior: the throughputs remains high, and the end-to-end
delay is about half that of the standard protocol.


ACKNOWLEDGMENTS

This work was funded partially by the Information Society Technologies Programme of
the European Commission, Future and Emerging Technologies, under the IST-2001-
38113 MobileMAN project, and partially by the Italian Ministry for Education and Scien-
tific Research in the framework of the FIRB-VICOM project. The authors wish to express

<<

. 23
( 87 .)



>>