<<

. 19
( 87 .)



>>

livered are generated by either an ftp application, or a continuous bit rate (CBR) applica-
tion. In the former, case the TCP protocol is used at the transport layer, whereas in the lat-
ter case UDP is the transport protocol.


Table 3.2. Throughput (in Kbps) Estimated in the Indoor Scenario when all Three ftp Sessions are
Active
TCP1 TCP2 TCP3 Aggregate
Basic Access 42 29.5 57 128.5
RTS/CTS 34 27 48 109
87
3.4 EXPERIMENTAL ANALYSIS OF IEEE 802.11 AD HOC NETWORKS




Figure 3.12. Reference network scenario for the outdoor experiments.



We performed a preliminary set of experiments aimed at estimating the Tx_range in
the outdoor environment where the experiments were done. We used the following proce-
dure. We considered a single couple of stations, S1 and S2. Then, starting from zero, we
progressively increased the distance d(1, 2), between these two stations until they were no
longer able to exchange data. For each value of d(1,2), the ping application was used to
test the connectivity between the stations. By applying this procedure several times, we
obtained that the transmission range is on the order of 40 m. It is worth pointing out that
in a real environment, the value of TX_range is not constant. It is highly variable depend-
ing on several factors: weather conditions, hour of the day, place and time of the experi-
ment, and so on.
Then we performed several experiments with Session 1 and Session 2 simultaneously
active. In all experiments, the receiving station is always in the transmission range of its
transmitting station; that is, Station S2 (S4) is in the transmitting range of Station S1 (S3).
On the other hand, the distance d(2, 3) between the two couples of stations6 is variable.
Depending on the actual d(2, 3) value, the following situation can occur.

1. All stations are within the transmission range of each other (Type 1). This means
that in our testbed, the distance between any two stations must be less than 40 m.
2. Extreme case: the two sessions are far from each other (Type 2). In our testbed, this
is achieved by setting d(2, 3) > 90 m (i.e., more than twice the minimum transmis-
sion range size);.
3. Intermediate case 1 is obtained by setting d(2, 3) = 65 m (Type 3).
4. Intermediate case 2 is obtained by setting d(2, 3) = 15 m (Type 4).

In all experiments ftp data traffic was transmitted and the TCP protocol was used at the
transport layer.7 For this reason, the two sessions will be indicated below as TCP1 and
TCP2. The payload size of TCP packets was set to 512 bytes.
The results obtained for Type 1 and Type 2 experiments are summarized in Table 3.3.
These experiments produced the expected results. In Type 1 experiments (all stations
within the same transmission range), the two ftp sessions fairly share the bandwidth, and
the aggregate throughput is close to the reference throughput for this configuration (see
Table 3.1). From the above results, it also appears that the RTS/CTS mechanism is useless
since it only reduces the aggregate throughput (due to the overhead introduced by the RTS
and CTS frames).

6
That is, the couple (3, 4) with respect to the couple (1, 2), and vice versa.
7
The length of each experiment is 120 seconds.
88 IEEE 802.11 AD HOC NETWORKS: PROTOCOLS, PERFORMANCE, AND OPEN ISSUES


Table 3.3. Throughputs in Kbytes/sec (Kbps) Measured in Type 1 and Type 2 Experiments
Type 1 Type 2
TCP 1 TCP 2 TCP 1 TCP 2
No RTS/CTS 61 54 122.5 122
RTS/CTS 59.5 49.5 96 100



In Type 2 measurements, the two sessions are independent, and they both achieve a
throughput very close to the reference throughput. Again, the RTS/CTS mechanism is
useless since it only introduces overhead.
Unlike the previous ones, Type 3 and Type 4 experiments exhibited a very strange and
unpredictable behavior, as shown in Figure 3.13 and Figure 3.14. In Type 3 experiments,
stations S2 and S3 are 65 m apart from each other. It can be observed that the use of the
RTS/CTS mechanism produces a capture of the channel by the second session (i.e.,
S3“S4). A possible explanation for this behavior is that Station S2 is often blocked by S3
data transmissions to S4. Hence, it may not be able to reply to the RTS frame of S1. On
the other hand, session S3“S4 is only marginally affected by session S1“S2, as the only
possible impact is due to S3 being blocked by S2™s (CTS and ACK) transmissions. When
using the basic access mechanism, S1 can start transmitting to S2 without almost any in-
terference from session S3“S4.
It is also worth noting that by using the basic access, the second session does not re-
duce its throughput (actually, the throughput of TCP2 increases as the RTS/CTS overhead
is removed). Indeed, with the basic access each session achieves a higher throughput.
To summarize, in this configuration the RTS/CTS mechanism, adding further correla-
tions between the stations™ behavior (S1 cannot start transmitting if S2 does not reply with
a CTS frame), produces a block of the first session without providing any advantage to the
other one.
In Type 4 experiments, whose results are shown in Figure 3.14, we observed the cap-
ture of the channel by one of the two TCP connections. In this case, the RTS/CTS mecha-
nism provided a little help in solving the problem.
The experimental results presented above confirm the unfairness/capture problems of
the TCP protocol in IEEE 802.11 ad hoc networks revealed in previous simulation studies.
As briefly discussed in Section 3.3, the TCP protocol (specifically the flow/congestion
control mechanism), by introducing correlations in the transmitted traffic, emphasizes
these phenomena. This effect is clearly pointed out by the experimental results shown in
Figure 3.15. This figure still refers to the Type 4 configuration but traffic flows are now
generated by CBR sources and the UDP protocol is used instead of TCP. As is clearly
shown, the capture effects disappear.
In conclusion, the experimental results have confirmed that, in some scenarios, TCP
connections may actually experience significant throughput unfairness, and even capture
of the channel by one of the connections, as pointed out in previous simulation studies.
Furthermore, it has been clearly shown that the RTS/CTS mechanism might be complete-
ly ineffective when there are stations that are outside their respective transmission ranges
but within the same carrier sensing range. In these cases, the physical carrier sensing is
sufficient to regulate the channel access and the virtual carrier sensing (i.e., the RTS/CTS
mechanism) is useless.
89
3.4 EXPERIMENTAL ANALYSIS OF IEEE 802.11 AD HOC NETWORKS


NO RTS/CTS


140


120


100
Throughput [Kbps]




80
TCP1
TCP2
60


40


20


0
#1 #2 #3



RTS/CTS


140


120


100
Throughput [Kbps]




80
TCP1
TCP2
60


40


20


0
#1 #2 #3

Figure 3.13. Throughputs (in Kbps) measured in the outdoor scenario in Type 3 experiments with
the RTS/CTS mechanism disabled (top) and enabled (bottom).
90 IEEE 802.11 AD HOC NETWORKS: PROTOCOLS, PERFORMANCE, AND OPEN ISSUES


NO RTS/CTS

140


120


100
Throughput [Kbps]



80
TCP1
TCP2
60


40


20


0
#1 #2 #3




RTS/CTS

140


120


100
Throughput [Kbps]




80
TCP1
TCP2
60


40


20


0
#1 #2 #3

Figure 3.14. Throughputs (in Kbps) measured in the outdoor scenario in Type 4 experiments with
the RTS/CTS mechanism disabled (top) and enabled (bottom).
91
3.4 EXPERIMENTAL ANALYSIS OF IEEE 802.11 AD HOC NETWORKS


NO RTS/CTS

140


120


100
Throughput [Kbps]




80
UDP1
TCP2
60


40


20

<<

. 19
( 87 .)



>>