Performance for Delay-Tolerant Data Collection

To evaluate the performance of the aggregation protocols for delay-tolerant data collection, a sufficiently large τ is chosen so that network contention is minimal. Maximum data age is not a major concern in delay-tolerant applications. Perfor- mance comparison between unicast and broadcast-based aggregation is focused on the end-to-end loss rate and traffic volume.

4.3.3.1 Principal Performance Comparison

Figures in this section show performance results for the sensor network with 160 nodes over a 250 meter by 250 meter area. Fig. 4.25 plots the end-to-end loss rate of the synchronous protocols, for different physical layer loss rates. The figure is

for fixed packet size; the figure for increasing packet size looks very similar and is omitted. Packet loss due to contention is rare when τ is sufficiently large. Packet size does not have much impact on end-to-end loss rate and the average number of MAC layer packet transmissions per round in this case. The average number of bytes transmitted per round, however, is still greatly influenced by packet size.

0.1% 1% 5% 10% 20% 40% 60% 80% 100% 0 0.2 0.4 0.6 0.8 1

end-to-end loss rate

physical layer loss rate unicast synch., 3X unicast synch., 4X unicast synch., 8X broadcast synch.

Figure 4.25: Performance of Synchronous Aggregation (delay-tolerant, N = 160, S = 250m × 250m, fixed packet size)

As seen in Fig. 4.25, synchronous broadcast-based aggregation outperforms syn- chronous unicast-based aggregation with up to 3 MAC layer retransmissions when physical layer loss rate is higher than 20%. It also outperforms synchronous unicast- based aggregation with up to 4 MAC layer retransmissions when physical layer loss rate is higher than 40%. However, retransmission is effective in packet recovery when packet loss is uniformly distributed. Synchronous unicast-based aggregation is able to yield lower end-to-end loss rate for different physical layer loss rates when up to 8 MAC layer retransmissions are allowed.

Fig. 4.26 shows the performance of the asynchronous protocols for different physical layer loss rates. As before, the figure is for fixed packet size; the figure for increasing packet looks very similar and is omitted. Asynchronous unicast-based aggregation with up to 8 MAC layer retransmissions consistently outperforms asyn- chronous broadcast-based aggregation for different physical layer loss rates.

0.1% 1% 5% 10% 20% 40% 60% 80% 100% 0 0.2 0.4 0.6 0.8 1

end-to-end loss rate

physical layer loss rate unicast asynch., 3X unicast asynch., 4X unicast asynch., 8X broadcast asynch.

Figure 4.26: Performance of Asynchronous Aggregation (delay-tolerant, N = 160, S = 250m × 250m, fixed packet size)

4.3.3.2 Impact of Density

Simulations for this section are done with the two sensor fields with 120 and 240 nodes randomly deployed in a 250 meter by 250 meter area. Only results for fixed packet size are shown; results for increasing packet size are quite similar.

Fig. 4.27 and Fig. 4.28 show the performance of the synchronous aggregation protocols, for the sensor fields with 120 and 240 nodes, respectively. Synchronous broadcast-based aggregation relies on multi-path routing for reliability. Lower den- sity means that a node has fewer neighbors to receive its broadcasts and carry its data in their aggregates. Fig. 4.27 shows that synchronous broadcast-based aggregation is consistently outperformed by synchronous unicast-based aggregation with up to 4 and 8 MAC layer retransmissions for different physical layer loss rates. The relative performance of synchronous broadcast-based aggregation improves with higher den- sity. For the sensor field with 240 nodes, synchronous broadcast-based aggregation consistently outperforms synchronous unicast-based aggregation with up to 3 MAC layer retransmissions for different physical layer loss rates. It also outperforms syn- chronous unicast-based aggregation with up to 4 MAC layer retransmissions when the physical layer loss rate is higher than 30%.

0.1% 1% 5% 10% 20% 40% 60% 80% 100% 0 0.2 0.4 0.6 0.8 1

end-to-end loss rate

physical layer loss rate unicast synch., 3X unicast synch., 4X unicast synch., 8X broadcast synch.

Figure 4.27: Impact of Lower Density on Synchronous Aggregation (delay- tolerant, N = 120, S = 250m × 250m, fixed packet size)

0.1% 1% 5% 10% 20% 40% 60% 80% 100% 0 0.2 0.4 0.6 0.8 1

end-to-end loss rate

physical layer loss rate unicast synch., 3X unicast synch., 4X unicast synch., 8X broadcast synch.

Figure 4.28: Impact of Higher Density on Synchronous Aggregation (delay- tolerant, N = 240, S = 250m × 250m, fixed packet size)

protocols, for the sensor fields with 120 and 240 nodes, respectively. Similar to synchronous broadcast-based aggregation, the relative performance of asynchronous broadcast-based aggregation improves as the network density increases.

0.1% 1% 5% 10% 20% 40% 60% 80% 100% 0 0.2 0.4 0.6 0.8 1

end-to-end loss rate

physical layer loss rate unicast asynch., 3X unicast asynch., 4X unicast asynch., 8X broadcast asynch.

Figure 4.29: Impact of Lower Density on Asynchronous Aggregation (delay- tolerant, N = 120, S = 250m × 250m, fixed packet size)

0.1% 1% 5% 10% 20% 40% 60% 80% 100% 0 0.2 0.4 0.6 0.8 1

end-to-end loss rate

physical layer loss rate unicast asynch., 3X unicast asynch., 4X unicast asynch., 8X broadcast asynch.

Figure 4.30: Impact of Higher Density on Asynchronous Aggregation (delay- tolerant, N = 240, S = 250m × 250m, fixed packet size)

4.3.3.3 Traffic Volume

Figures in this section show performance results for the sensor network with 160 nodes over a 250 meter by 250 meter area. Fig. 4.31 shows the average number of MAC layer packets that are transmitted per round with the synchronous aggregation

protocols, for fixed packet size. The figure for increasing packet size is very similar. As described before, each node broadcasts at most twice per round in broadcast- based aggregation. In unicast-based aggregation, at least two MAC layer packet transmissions are required for each successful unicast transmission. For all physical layer loss rates, synchronous broadcast-based aggregation sends fewer MAC layer packets per round than synchronous unicast-based aggregation.

0 200 400 600 800 1000 1200 1400 1600 0 0.2 0.4 0.6 0.8 1

MAC layer packet transmissions per round

physical layer loss rate unicast synch., 3X

unicast synch., 4X unicast synch., 8X broadcast synch.

Figure 4.31: MAC layer Packet Transmissions per Round of Synchronous Aggregation (delay-tolerant, N = 160, S = 250m × 250m, fixed packet size)

0 10000 20000 30000 40000 50000 60000 70000 80000 90000 0 0.2 0.4 0.6 0.8 1

bytes transmitted per round

physical layer loss rate unicast synch., 3X

unicast synch., 4X unicast synch., 8X broadcast synch.

(a) Increasing Packet Size

0 10000 20000 30000 40000 50000 60000 70000 80000 90000 0 0.2 0.4 0.6 0.8 1

bytes transmitted per round

physical layer loss rate unicast synch., 3X

unicast synch., 4X unicast synch., 8X broadcast synch.

(b) Fixed Packet Size

Figure 4.32: Bytes Transmitted per Round of Synchronous Aggregation (delay-tolerant, N = 160, S = 250m × 250m)

Fig. 4.32 shows the average number of bytes that are transmitted per round with the synchronous aggregation protocols. For increasing packet size, the data volume with synchronous broadcast-based aggregation is larger than with the other protocols when the physical layer loss rate is lower than 40%, owing to cases in which

the same sensor value is redundantly included in multiple aggregates. Note that the data volume with this protocol keeps decreasing as the loss probability increases, since the growing packet loss decreases this redundancy. For fixed packet size, the traffic volume with synchronous broadcast-based aggregation is consistently lower for all loss rates.

Fig. 4.33 shows the average number of MAC layer packets that are transmitted per round with the asynchronous aggregation protocols, for fixed packet size. The figure for increasing packet size is very similar. Fig. 4.34 plots the average number of bytes that are transmitted per round with the asynchronous aggregation protocols, for both fixed and increasing packet size. The results are quite similar with those for synchronous aggregation.

4.3.3.4 Impact of Two-State Gilbert Error Model

Broadcast and unicast-based aggregation using the two-state Gilbert error model are evaluated in this section. The proportion of time each node spends in the bad state is fixed at 20%. The average sojourn time in the bad state is varied to model a range of scenarios. With very short sojourn times, independent random packet loss is modelled. With long sojourn times, link outages and partial node failures are modelled.

Fig. 4.35 and Fig. 4.36 plot performance results of the synchronous and asyn- chronous aggregation protocols, respectively, for the sensor field with 160 nodes over a 250 meter by 250 meter area. Only results for fixed packet size are shown; results for increasing packet size are very similar. For long average sojourn time in the bad state, broadcast-based aggregation yields much lower end-to-end loss rate than the unicast-based aggregation protocols, because MAC layer retransmissions are not effective in packet recovery in this case. As seen in Fig. 4.35 and Fig. 4.36, both syn- chronous and asynchronous unicast-based aggregation protocols yield around 40% end-to-end loss rate with long average sojourn time in the bad state (2 seconds, for example). For a very small average sojourn time, such as 0.0001 seconds in the fig- ures, the end-to-end loss rate increases, since a node is likely to enter the bad state

0 200 400 600 800 1000 1200 1400 1600 0 0.2 0.4 0.6 0.8 1

MAC layer packet transmissions per round

physical layer loss rate unicast asynch., 3X unicast asynch., 4X unicast asynch., 8X broadcast asynch.

Figure 4.33: MAC layer Packet Transmissions per Round of Asynchronous Aggregation (delay-tolerant, N = 160, S = 250m × 250m, fixed packet size)

0 10000 20000 30000 40000 50000 60000 70000 80000 90000 0 0.2 0.4 0.6 0.8 1

bytes transmitted per round

physical layer loss rate unicast asynch., 3X unicast asynch., 4X unicast asynch., 8X broadcast asynch.

(a) Increasing Packet Size

0 10000 20000 30000 40000 50000 60000 70000 80000 90000 0 0.2 0.4 0.6 0.8 1

bytes transmitted per round

physical layer loss rate unicast asynch., 3X unicast asynch., 4X unicast asynch., 8X broadcast asynch.

(b) Fixed Packet Size

Figure 4.34: Bytes Transmitted per Round of Asynchronous Aggregation (delay-tolerant, N = 160, S = 250m × 250m)

at least once while receiving a packet, causing the loss of that packet. 0.1% 1% 5% 10% 20% 40% 60% 80% 100% 0.0001 0.001 0.01 0.1 1 5

end-to-end loss rate

average duration of bad state (in seconds) unicast synch., 3X

unicast synch., 4X unicast synch., 8X broadcast synch.

Figure 4.35: Performance of Synchronous Aggregation for Two-state Gilbert Error Model (delay-tolerant, N = 160, S = 250m × 250m, Pb = 20%, fixed

packet size) 0.1% 1% 5% 10% 20% 40% 60% 80% 100% 0.0001 0.001 0.01 0.1 1 5

end-to-end loss rate

average duration of bad state (in seconds) unicast synch., 3X unicast synch., 4X unicast synch., 8X broadcast synch.

Figure 4.36: Performance of Asynchronous Aggregation for Two-state Gilbert Error Model (delay-tolerant, N = 160, S = 250m × 250m, Pb = 20%, fixed

packet size)

4.3.3.5 Broadcast-based Synchronous vs. Broadcast-based Asynchronous Fig. 4.37 shows that broadcast-based synchronous and asynchronous aggregation protocols yield similar performance for different physical layer loss rates. The main reason to use asynchronous timing control is to lower packet forwarding delay at

intermediate nodes. The broadcast-based asynchronous protocol does not have much advantage in the delay-tolerant scenario.

0.1% 1% 5% 10% 20% 40% 60% 80% 100% 0 0.2 0.4 0.6 0.8 1

end-to-end loss rate

physical layer loss rate broadcast synch. broadcast asynch.

(a) Increasing Packet Size

0.1% 1% 5% 10% 20% 40% 60% 80% 100% 0 0.2 0.4 0.6 0.8 1

end-to-end loss rate

physical layer loss rate broadcast synch. broadcast asynch.

(b) Fixed Packet Size

Figure 4.37: Broadcast-based Synchronous vs. Broadcast-based Asyn- chronous Aggregation (delay-tolerant, varying τ , N = 160, S = 250m × 250m)

4.3.3.6 Summary of Results

Section 4.3.3 evaluates the performance of the protocols in the context of delay- tolerant data collection. Sections 4.3.3.1 to 4.3.3.3 present performance results with the independent random error model. The results show that when packet loss is random, the unicast-based protocols are able to improve reliability by increasing the maximum number of MAC layer retransmissions. The broadcast-based protocols relies on multipath routing for reliability, and their relative performance improves with higher density. The broadcast-based protocols transmit fewer packets per round than the unicast-based protocols. For fixed packet size, the broadcast-based protocols also transmit fewer bytes per round than the unicast-based protocols. For increas- ing packet size, traffic volume with broadcast-based aggregation decreases as the physical layer loss rate get higher, while more traffic is generated by unicast-based aggregation as the physical layer loss rate increases. Performance results with the two-state Gilbert error model are presented in Section 4.3.3.4. Both synchronous and asynchronous broadcast-based aggregation yield significantly improved reliabil- ity than their unicast-based counterparts, because MAC layer transmission is not

effective in loss recovery in this case.