Check out the new USENIX Web site. next up previous
: Conclusions and Future Work : Experimental Results : Beacon delays in contention-free


Bandwidth Estimation

In a collision-free environment, we know from Section 4.1 that the mean beacon delay is 552 $\mu s$. For a packet of size $L$ bytes and data rate $R$, the potential downstream bandwidth is then given by (Eq. 4):

\begin{displaymath}B = \frac{8L}{552 + \frac{8L}{R} + T_{A}} \end{displaymath}

where $T_{A} = 213 \mu s$. For instance, when $L=640$ and $R=$11 Mbps, the potential downstream bandwidth yields an estimate $B =$ 4.16 Mbps.

We performed a simple experiment to verify whether the actual bandwidth observed on the downlink from AP to an end-host compares with the estimated value obtained above. A UDP session is initiated from the AP to an affiliated end-host. The duration of the transfer was 200 seconds and the AP was constantly backlogged. The actual bandwidth $B_{m}$ from the AP to the end-host for the duration of the transfer was measured to be 4.3 Mbps, which closely agrees with the estimate $B$ obtained above.

Figure 3: Beacon Delays when the AP is loaded
\begin{figure}\begin{center}
\input{epsf}\setlength{\epsfxsize}{2.4in}
\epsfbox{beacon_delay_1.eps}
\vspace*{-2mm}
\vspace*{-6mm}
\end{center}\end{figure}

In a second experiment, we place one AP and two wireless hosts $H1$ and $H2$ in the anechoic chamber. Host $H1$ is affiliated to the AP. A UDP session is initiated from the AP to the host $H1$. The UDP session consists of CBR traffic generated at the rate of 100 packets/second, each packet of size 576 bytes (640 bytes including the headers). Another host $H2$ is configured in ``monitor'' mode and records the delays observed for the beacon frames (beacons numbered 300 and higher in Figure 3). The monitoring host $H2$ estimates the mean beacon delay from the AP over the duration of the UDP transfer to be $687 \mu s$. Using our bandwidth estimation methodology, $H2$ estimates the potential bandwidth from the AP to itself to be $B =$3.74 Mbps. We then affiliate $H2$ with the AP and initiate another UDP session between the AP and host $H2$, simultaneously with the UDP session between the AP and $H1$. The AP is always backlogged with packets for $H2$. The actual bandwidth $B_{m}$ from the AP to $H2$ is measured to be 4.06 Mbps, which agrees with the potential bandwidth estimate of 3.74 Mbps, obtained by $H2$ prior to affiliation with the AP. Thus, the experimental results suggest that our approach is promising.


next up previous
: Conclusions and Future Work : Experimental Results : Beacon delays in contention-free