Energy Efficient and Reliable Networking
6.4 Increasing the Reliability
From the previous examples, it was clear that CICADA experiences a low packet loss. Although, in a more dynamic environment, a higher packet loss can occur due to interference or the way the tree is set up. Therefore, we will add more re-liability mechanisms to the protocol. Three adaptations are envisioned: the intro-duction of an acknowledgment map, randomization of the schemes and repeating the SCHEME-messages of the parent (aka overhearing).
6.4.1 Acknowledgments
The way acknowledgments are used is similar to that of WASP (see Section 5.3.1.6).
At the end of each SCHEME-message, an ACK-sequence for the previous cycle is sent: it contains a bit for each slot in which data was sent. A 0 denotes that the packet was not received correctly, a 1 denotes success. The node the SCHEME-message is destined for, say node i, will check the ACK-sequence. If the position of a 0 corresponds to one of the slots where i sent data, the node will resend that data in this cycle. Its parent will already include an extra slot for node i by ad-justing it’s αi and βi. For example, in the network of Figure 6.1 the sink S adds the following ACK-sequence to it’s scheme: 11101. Nodes A and B receive these messages and node B knows that it has sent in the fourth data slot. Node B will retransmit that packet and delete the other packets from its buffer.
2The implementation is mainly done by Wim Torfs of the University of Antwerp
CHAPTER6 157
6.4.2 Randomization of the Schemes
By randomization of the schemes, we try to restrict the influence of interference.
When the tree is set up, it might happen that two nodes i and j can hear each other, but have different parents, k and l respectively, e.g. because the link between k and i is more reliable than the link between l and i. When the schemes are fixed, i.e. the first node to join always sends first and so on, it might well happen that nodes i and j will interfere while sending their data to their respective parents. By randomizing the schemes, i.e. by changing the sequence in which the children are allowed to send, the overall interference will be decreased.
6.4.3 Overhearing
During our simulations we noticed that, from time to time, nodes miss a SCHEME-message from their parent, because of a link that’s not very stable. The result is that this node and all nodes below it cannot do anything and must have their radio on until the next cycle. In order to tackle this problem, a child node repeats the scheme of its parent when it sends its own scheme. Doing so, siblings can exploit this information if they missed it.
6.4.4 Analysis
The path loss model (2.4), the link probability (2.12) and the improvements are implemented. The simulator was used to analyze the changes to the protocol. The size of the network is varied from 5 to 30 nodes and the nodes were randomly placed in a 2 by 2 meter area with the sink positioned in the center. The distances between the nodes is at most 40 cm in a connected topology, i.e. every node is within transmission range of at least one other node so there is always a path to the sink. Nodes start randomly, they do not join the network all at once. All simula-tions were run during 10.000 slots for 1000 randomly generated topologies, while making sure the same topologies are used in comparisons. Each node generates one packet after a fixed period. This period is defined in slots and equals three times the number of nodes in the network. The number of packets received by the sink and the number of retransmissions are considered. We also look at the number of slots the radio was on, averaged out over the number of runs, to study the impact on network lifetime. The randomization and overhearing mechanism are evaluated separately and combined. The ACK-mechanism is always used.
Results of the simulations are shown in Figure 6.18(a). The values represent the improvement in percentage between the results without and with randomiza-tion. It can be seen that scheme randomization has a positive impact. The number of packets that can be received by the sink increases by more than 4% for larger networks. This is caused by the lower number of collisions. Yet, it can be seen that
158 CICADA: ENERGYEFFICIENT ANDRELIABLENETWORKING
Figure 6.18: Evaluation of the proposed reliability mechanisms. The figure plots the dif-ference between regular CICADA and CICADA with the proposed reliability mechanism. For example, in (a) the reliability mechanism has 5% more re-ceived packets than regular CICADA for a network of 20 nodes.
the number of retransmissions is larger. The rise of the number of retransmissions, however, is lower than the rise of the number of received packets. Thus, relatively spoken, the number of retransmissions has not increased. The absolute increase can be explained by the higher number of transmissions in the network. For small networks, little effect was found as the parent nodes have few child nodes to ran-domize. It is important to notice that the figure also shows that the average time a radio is on is almost similar with or without randomization. This is expected as the number of slots a node sleeps does not change. Hence, the scheme randomization leads to an higher throughput in the system while having almost no impact on the network lifetime.
The results for the overhearing mechanism are shown in Figure 6.18(b). We see that the sink receives about 10% more data. The chance of missing the parent’s scheme is lower as the scheme can be recovered by listening to the sibling’s control packets. The node now knows when to send its data, which will increase the throughput. The number of retransmissions drops dramatically with roughly 70-80%. This shows that overhearing has a positive effect as less control messages are missed and the nodes thus know when to send their data, leading to fewer retransmissions. The impact on the energy consumption is very low.
The overhearing solution also increases the overhead. When the scheme of the parent is included to the SCHEME-message, an additional of 4 + 2 · xpbytes are added, where xpindicates the number of children of the parent. If we assume that in a network each node has a maximum of 10 children, the length of the control packet will change from 25 bytes to 49 bytes. This means that the length of the slot size in the control cycle needs to be increased, which will have an impact on the energy efficiency. However, these influences are minor as long as the slot ratio
CHAPTER6 159
is larger than ten, as was shown in Section 6.2.2. Hence, if a data slot can hold a message of 500 bytes, the influence of adding your parent’s scheme is minor.
5 10 15 20 25 30
(a) Overhearing with low packet interarrival time
5 10 15 20 25 30
Figure 6.19: Evaluation of the proposed reliability mechanisms (2).
We have also evaluated the influence of a low packet inter arrival time. Each node now sends data after a period of 1.5 times the number of nodes in the network.
For smaller networks, this is higher than the duration of the cycle, for larger net-works not. Results can be seen in Figure 6.19(a). The number of retransmissions caused by the overhearing mechanism drops for larger networks. The network is already congested and the overhearing mechanism only gives a limited solution for a congested network.
The combined effect of these two mechanisms are shown in Figure 6.19(b) for high packet inter arrival time. It can be seen that the effects of the overhearing mechanism dominates. The combined solution further has little influence on the energy performance.