<<

. 18
( 87 .)



>>



3.3.4 Effects of the Interaction between MAC Protocol
and TCP Mechanisms
The interaction of some features of the 802.11 MAC protocol (hidden-/exposed-station
problem, exponential backoff scheme, etc.) with the TCP protocol mechanisms (mainly,
the congestion control mechanism) may lead to several unexpected and serious problems.
S. Xu and Saadawi identified these problems through a simulation analysis of a multihop
ad hoc network via the ns network simulator tool [26]. The same results have been con-
firmed with a different simulation tool [27]. Recently, similar phenomena have been also
observed in other scenarios [27, 28].
Specifically, in [26] and [27] it is pointed that the following problems may affect the
TCP performance in a multihop ad hoc environment:

1. The instantaneous throughput of a TCP connection may be very unstable (dropping
frequently to zero), even when this is the only active connection in the network (in-
stability problem).
2. In case of two simultaneous TCP connections, it may happen that the two connec-
tions can not coexist: when one connection develops, the other one is shut down (in-
compatibility problem).
3. With two simultaneous TCP connections, if one connection is single-hop and the
other one is multiple-hop, it may happen that the instantaneous throughput of the
multiple-hop connection is shut down as soon as the other connection becomes ac-
tive (even if the multiple-hop connection starts first). There is no chance for the
multiple-hop connection once the one-hop connection has started (one-hop unfair-
ness problem).

The above problems have been revealed in a string network topology like the one
shown in Figure 3.8, where the distance between any two neighboring stations is 200 m
and stations are static. According to the 802.11-based Wave-Lan, the nominal transmis-
sion radius of each station has been set to 250 m (each station can thus communicate only
82 IEEE 802.11 AD HOC NETWORKS: PROTOCOLS, PERFORMANCE, AND OPEN ISSUES


with its neighboring stations). Furthermore, the sensing and interfering ranges have been
set to twice the transmission range of 500 m [26, 27], which is the typical setting of the
ns-2 simulator.
Below we will provide a brief explanation of how the one-hop unfairness problem aris-
es. Similar explanations can be provided for the instability and incompatibility problems,
but are omitted for the sake of space. The reader can refer to [27] for a detailed analysis of
all cases.
Figure 3.9 shows two TCP connections. The first connection is from Station 2 to Sta-
tion 3 (one-hop connection), whereas the second connection is from Station 6 to Station 4
(two-hop connection). Let us assume, for example, that Station 2 is transmitting a data
frame to Station 3 (e.g., a TCP segment), and Station 5 wants to transmit a frame to Sta-
tion 4. According to the 802.11 MAC protocol, Station 5 tries to send an RTS frame, and
then waits for the corresponding CTS frame. However, Station 5 never receives this CTS
frame.
Most of the RTS transmission attempts tried by Station 5 result in a collision at Station
4 due to the interference of Station 2 (hidden-station problem). Station 5 cannot hear the
CTS from Station 3 because it is out of the transmission range of Station 3 and, thus, it is
not aware of Station 2 transmission. However, Station 4 is in the interfering range of Sta-
tion 2 since the interfering range is larger than the transmission range (twice in ns-2 simu-
lator). Even if Station 4 successfully receives the RTS frame, it is not able to reply with
the corresponding CTS frame, again due to Station 2. Though Station 4 is out of the trans-
mission range of Station 2, Station 4 can sense the transmission of Station 2 since the
sensing range is larger than the transmission range (twice in the ns-2 simulator). This in-
hibits Station 4 from accessing the wireless medium (exposed-station problem).
After failing to receive the CTS frame from Station 4 seven times, Station 5 reports a
link breakage to its upper layer and a route-failure notification is sent to Station 6 (the
data packet originator). Upon receiving this notification, Station 6 starts a route discovery
process. Obviously, while looking for a new route no data packet can flow along the con-
nection and this makes the instantaneous throughput drop to zero.
The above example allows us to understand why the instantaneous throughput of the
two-hop connection drops to zero. However, it not yet clear why this throughput remains
at zero for most of the connection lifetime. To better clarify this point, the following addi-
tional remarks need to be taken into account.

Since Station 5 is in the interfering range of Station 3, it has to defer when Station 3
is sending. Therefore, Station 5 can transmit an RTS frame only when Station 3 is
not sending.




Figure 3.9. A string topology with two TCP connections. The first connection is one-hop; the sec-
ond connection is two-hop.
83
3.4 EXPERIMENTAL ANALYSIS OF IEEE 802.11 AD HOC NETWORKS


In the one-hop connection, as soon as Station 2 receives a TCP ACK from Station 3,
it immediately prepares itself to send another TCP segment. This means that Station
5 has very few opportunities to find the channel free.
Data frames (i.e., TCP segments) transmitted by Station 2 are usually much larger in
size than the RTS frames that Station 5 tries to transmit.

In conclusion, the time available for Station 5 for successfully accessing the channel is
very small. In addition, the exponential backoff scheme used by the 802.11 MAC protocol
always favors the last succeeding station.
From the above description, it emerges that several features of the multihop ad hoc en-
vironment contribute to the “capture” of the channel by the one-hop connection. The most
important and direct causes are the hidden-station and the exposed-station problems.
These problems, in turn, are caused by the larger size of the interfering and sensing ranges
with respect to the transmission range. However, the random backoff scheme of the
802.11 MAC protocol also contributes by favoring the last succeeding station.
The “capture” effect revealed in [26, 27] is not peculiar to the string network topology.
Gerla et al. observed the same phenomenon even in other scenarios [28, 29]. In [28], they
also proposed two possible solutions to remove the capture effect: (i) replacement of the
binary backoff scheme in the 802.11 MAC protocol by an adaptive retransmission timeout
based on the number of active neighboring stations; and (ii) the use of special antennas
that reduce interference during packet reception.


3.4 EXPERIMENTAL ANALYSIS OF IEEE 802.11 AD HOC NETWORKS

In the previous section, we have seen that there exists an extensive literature that has inves-
tigated TCP performance in ad hoc networks, especially over the IEEE 802.11 MAC proto-
col. Most papers report the same type of unfairness problems. The hidden- and exposed-
station problem, the large interference range, and the backoff scheme of IEEE 802.11 MAC
protocol have been recognized as the major reasons for these unfairness problems. All these
previous analyses were carried by simulation and, hence, the results observed are highly de-
pendent on the physical layer model implemented in the simulation tool used in the analy-
sis (e.g., GloMosim [11], ns-2 [20], Qualnet [23]). Hereafter, we validate and extend these
results by presenting a similar analysis that has been carried on a real testbed. Since the sim-
ulation results presented in Section 3 were obtained by considering IEEE 802.11 network
cards operating at the nominal bit rate of 2 Mbps, most of the measurement studies pre-
sented in this section refer to the IEEE 802.11 standard [15]. However, in Section 3.5 we
will also investigate the performance of the IEEE 802.11b ad hoc networks.
It is worth pointing out that, although in the simulation studies presented above the val-
ues of TX_range, PCS_range, and IF_range are known and constant, in the real world the
physical channel has time-varying and asymmetric propagation properties. Hence, the
values of TX_range, PCS_range, and IF_range may be highly variable, even during the
same experiment.

3.4.1 Experimental Testbed
The measurement testbed is based on laptops running the Linux-Mandrake 7.2 operating
system. The laptops are equipped with Lucent WaveLAN IEEE 802.11 network cards us-
84 IEEE 802.11 AD HOC NETWORKS: PROTOCOLS, PERFORMANCE, AND OPEN ISSUES


ing the DSSS technique, and operating at the nominal bit rate of 2 Mbps. The target of our
study is the analysis of the TCP performance over an IEEE 802.11 ad hoc network. Since
the aim of the study is to investigate the impact of the CSMA/CA protocol on the TCP
performance, static ad hoc networks (i.e., the network stations do not change their posi-
tion during an experiment) with single-hop TCP connections were considered. This allows
one to remove other possible causes that may interfere with the TCP behavior, such as,
link breakage, route recomputation, etc.

3.4.2 Indoor Experiments
The indoor experiments were carried out in the scenario depicted in Figure 3.10. Stations
numbered as S1, S2, and S3 have an active ftp session toward Station S4; that is, data
frames are transmitted to S4, which replies with ACK packets. As ftp data transfers are
supported by the TCP protocol, in the following the data flows will be denoted as TCPi,
where i is the index of the transmitting station. As shown in the figure, a reinforced con-
crete wall (represented by the gray rectangle) is located between stations S1 and S2, and
between stations S2 and S3. As a consequence, S1, S2, and S3 are outside the TX_ range
of each other.5 Furthermore, each Station Si (where i = {1, 2, 3}) is in the transmission
range of S4. Therefore, this is a typical hidden-station scenario in which it is expected that
the RTS/CTS mechanism (by avoiding hidden-station collisions) should provide a signifi-
cant throughput gain with respect to the basic CSMA/CA protocol.
Two sets of experiments were performed in this scenario. In the first set, only two ftp
sessions are active: TCP1 and TCP2. In the second set, all three sessions are active.
To better analyze the results, we also performed some reference experiments. Specifi-
cally, we measured the maximum throughput (at the application layer) of a single
sender“receiver session when the two stations are very close to each other (in the same
room), and no other session is active. The estimated throughput represents the upper
bound throughput for a sender“receiver session and is reported in Table 3.1 for different
operating conditions.
Let us now start analyzing the results related to the indoor scenario. The results ob-
tained in the scenario with two active sessions (TCP1 and TCP2) are summarized in Fig-
ure 3.11. These results refer to a 60 second ftp transfer that utilizes TCP packets with a
1460 byte payload size. Two types of experiments were done: with and without the
RTS/CTS mechanism. For each type, we performed three experiments under the same
conditions.
The following remarks can be made based on the above results:

1. The RTS/CTS mechanism does not provide any significant performance improve-
ment with respect to the basic access mechanism.
2. The RTS/CTS mechanism provides an aggregate throughput slightly lower than the
basic access mechanism. This is due to the additional overhead introduced by the
RTS and CTS frames.
3. In each experiment, the aggregate throughput is not very far from the reference
throughput reported in Table 3.1 (i.e., 145 and 135 Kbps for the basic access and the
RTS/CTS mechanism, respectively).

5
This was verified by running the Ping program for a sufficiently long time from each station to the other sta-
tions. In no case was a packet successfully delivered among each couple of stations.
85
3.4 EXPERIMENTAL ANALYSIS OF IEEE 802.11 AD HOC NETWORKS




Figure 3.10. Indoor scenario.



These observations are confirmed by the results obtained in the scenario with three ftp
sessions active, summarized in Table 3.2. For each set of experiments, Table 3.2 reports
the throughput averaged on all the experiments performed under the same conditions.
These results indicate that the carrier sensing mechanism is still effective even if the
transmitting stations are “apparently” hidden from each other. This can be explained by
remembering that the carrier sensing range is about twice the transmission range. Hence,
if two stations (outside the transmission range of each other) are in the transmission range
of a third station, there is a very high probability that they can sense each other. In these
cases, the physical carrier sensing is effective and, hence, adding virtual carrier sensing
(i.e., the RTS/CTS mechanism) is useless.

3.4.3 Outdoor Experiments
To better investigate the phenomena observed in the indoor environment, the testbed was
moved to an outdoor space. Each station was located in an open environment (a field
without buildings) in order to analyze the TCP behavior when hidden and/or exposed sta-
tions may be present. In all experiments, the WLAN was set to 2 Mbps.
The network scenario for the outdoor experiments is shown in Figure 3.12. In this sce-
nario, we may have two contemporary active sessions. Specifically, Station S1 communi-


Table 3.1. Reference Throughputs in Kbytes/sec (Kbps)
Packet size Packet size
1460 Bytes 512 Bytes
ftp/TCP traffic ftp/TCP traffic CBR/UDP traffic
Throughputs Basic Access 145 Kbps 125 Kbps 165 Kbps
Throughputs RTS/CTS 135 Kbps 110 Kbps 140 Kbps
86 IEEE 802.11 AD HOC NETWORKS: PROTOCOLS, PERFORMANCE, AND OPEN ISSUES


NO RTS/CTS
NO RTS/CTS

100
90
80
70
Throughput [Kbps] 60 TCP1
TCP1
50 TCP2
TCP2
40
30
20
10
0
#1 #2 #3


RTS/CTS
RTS/CTS


100
90
80
Throughput [Kbps]




70
TCP1
60
TCP2
50
40
30
20
10
0
#1 #2 #3


Figure 3.11. Throughput (in Kbps) estimated in the indoor scenario with two ftp sessions with the
basic CSMA/CA access (left) and the RTS/CTS mechanism (right), respectively.



cates with Station S2 (Session 1), while Station S3 is in communication with Station S4
(Session 2). In the figure, the arrows represent the direction of the data flow (e.g., S1 is
delivering data to S2), and d(i, j) is the distance between stations Si and Sj. Data to be de-

<<

. 18
( 87 .)



>>