Simulation results

We compared the above mentioned schedulers among different criteria, we

measured delay, throughput, and overhead for different network sizes, and the results are shown as follows:

5.2.1 Throughput

Total throughput VS Network size

0 0.5 1 1.5 2 2.5 3 3.5

4 8 12 16 20 24 28 32 36

Number of stations

Total Throughput

Reference APS

Round-robin

Figure 5.1: Total throughput against network size for Reference, APS, and RR schedulers

From Figure 5.1 we can notice a significant increase in the throughput for APS comparing to RR and reference schedulers. Reference scheduler starts at 0.5 Mbps when the network consists of 4 mobile nodes, so do for APS and RR. Throughput keeps increasing for all schedulers when the number of mobile nodes is 8, with a little difference for RR’s and APS’s account. The gap between APS, RR, on a hand and the reference on the other hand getting increased when the network size is getting larger, because reference scheduler keeps on the same throughput whatever the network size was, while the throughput of RR and APS is growing continuously as the network size increases. When the network size is larger than 28 nodes APS shows a much better performance than RR and the gap between them is getting bigger as the number of mobile nodes in the network increases.

Reference scheduler can support a limited number of transmission queues, which means that the number of supported voice stations is limited, when the network size increases, its performance doesn’t show any enhancement. On the contrary it is getting worse but in a very small percentage, and it is not noticeable. RR shows a good performance as the number of mobile stations increases but the APS shows a much better performance. The differentiation between the stations in the talking state and the stations in silence state, the ordered polling list according to the traffic directions and the TXOPs calculation method played an important role to make this difference.

5.2.2 Delay

Figure 5.2: Average delay against network size for Reference, APS, and RR schedulers Figure 5.2 shows the average transfer delay against different network sizes, it shows a significant growing gap between RR and reference schedulers on one side and APS scheduler on the other side. The higher delay in both RR and reference scheme is primarily due to the polling-induced overheads, in addition to the fairness in polling stations, which reaches 99.98% for the reference scheduler; while APS shows a very low

delay comparing to the other two schemes, this gap is explained by the unfairness and piggybacking features of the APS.

5.2.3 Packet loss ratio

Packet Loss Ratio VS Network size

0

Figure 5.3: Packet loss ratio against network size for Reference, APS, and RR schedulers

Figure 5.3 shows the packet loss ratio, it shows a good performance for APS comparing to the two other schemes. Voice packets are generated continuously every 20 ms in G711 codec, which means that those stations should be polled as soon as possible to give them the chance to send their voice packets, but in RR and reference schedulers, and with the increase of the network size, talking station may not be polled at the right time, thus, stations may start dropping packets from its TC, and this is the main reason of the packet loss ratio in RR and reference schedulers. On the other hand APS classifies stations into several types and give each type a priority to be polled, Table 4.2 shows the priorities given to the stations, talking stations with downlink will be at the top of the

priorities so the talking stations will always have the chance to send their packets, this will affect the packet drop ratio positively, and enhance the overall performance.

5.2.4 Polling Overhead analysis

We show analytically how polling overhead is decreased in APS comparing with RR. Our analysis inputs are: fixed TXOPs of 10 ms, Beacon interval of 100 ms, and 4 mobile nodes. In RR HC polls stations sequentially, that means it will poll silent stations as well as talking stations.

The polling frame consists of 36 bytes as mentioned in chapter 3. It should be sent at the base rate (2 Mbps), the PHY overhead is considered to be 192 µs. the total transmission time for the CF-Poll packet transmission is (36 * 8 / 2) + 192 = 336 µs. The 4 stations may be polled 2.5 times at maximum and not polled at all at minimum, so let’s say the average of the 4 stations to be polled is (0 + 2.5) / 2 = 1.25 times. For 4 station there will be 4 * 1.25 CF-Polls per Beacon interval which is equals 6 CF-Polls. The total transmission time for those 6 CF-Poll packets is 6 * 336 = 2016 µs per beacon interval. If we run the RR for 500 second, then the total CF-Poll transmission time is 2016 * 5 = 10080 µs = 10.08 ms. From the transmission time for the voice packet of G.711 codec is 196 * 8 / 11 = 142.55 µs, we can notice how many voice packets we can send during the polling overhead.

In APS we use piggybacking, and according to the way we order the polling lists it will be much more efficient to send a QoS Data + CF-Poll than sending only CF-Poll, especially that the piggybacked packet is considered to be a data packet, and it could be sent on the negotiated data rate between station and HC, and by doing the same calculations as we did for RR, we get (36 + 160) * 8 /4 = 392 µs to send one poll and one voice packet on the data rate of 4 Mbps, for 6 polls 392 * 6 = 2352 µs and for 5 beacon intervals 2352 * 5 = 11760 µs = 11.76 ms compared to 10.08 ms for RR but with the difference that we have sent 30 voice packets piggybacked with the Poll packets. Sending the same number of voice packets that have been sent piggybacked with CF-Polls in APS without piggybacking costs (160 * 8 / 11) * 6 * 5 = 3490.9 µs on 11 Mbps data rate. By subtracting the time required to transmit those voice packets without piggybacking them with CF-Polls 11760 – 3490.9 = 8269.1 µs = 8.3 ms.

Comparing the time required to transmit polling packets for APS and RR we see that APS decreased the polling overhead, especially with the growth of network size.

Based on the previous results we can notice a significant enhancement in APS comparing to the Reference and RR polling mechanisms. Table 5.2 shows the enhancement percentage of APS against RR and Reference.

Table 5.2: Percentage of enhancement for APS against RR and Reference.

APS VS Reference APS VS RR

Network

Size throughput Delay

Packet loss Ratio

throughput Delay

Packet loss Ratio

4 3.75% 78.57% 46.67% 2.89% 72.73% -1.82%

8 21.33% 90.23% 41.20% 11.08% 90.23% 30.51%

12 35.76% 86.70% 52.37% 5.67% 87.65% 50.93%

16 53.82% 73.22% 69.60% 11.17% 76.14% 65.49%

20 73.82% 78.03% 69.83% 13.69% 78.46% 68.10%

24 80.70% 80.11% 69.73% 15.57% 80.44% 69.73%

28 97.52% 80.49% 70.97% 18.60% 81.23% 73.63%

32 124.45% 82.45% 73.75% 22.88% 83.62% 78.92%

36 160.35% 82.18% 73.70% 40.42% 83.03% 79.66%