In both WSEDR and RTCF, EDR is adopted as the main metric for forwarding. EDR can precisely indicate which node is better to be chosen as the next-hop node, but it is not so easy and expensive to calculate EDR. In order to calculate the EDR, a node needs to gather the information of some neighbors which have the lower hop count. Because of the unreliable links, it is not enough for the sink node to only send one initialization packet. In order to have a preciser EDR, the sink node has to send more than one initialization packets. The more the initialization packets are sent, the preciser the EDR will be. However, the energy consumption will be higher as the number of the initialization packets increases. As a result, there exists a trade-off between the accuracy and the energy consumption. It is interesting to know the
6.2 : Future work 115
optimal number of initialization packets and to find other metrics that are easier to calculate and have the comparable performances.
The performances of all our proposed protocols are verified by simulations and mathemat- ical analysis. Although the simulation results can well show the performances of the proposed protocols, it is still necessary to verify our proposed protocols on a realistic testbed because the simulation models (e.g. physical layer models) used in the simulator can only reflect the realistic scenarios to a certain extent even if the used simulator, WSNET, can modelize a lot of physical phenomena (shadowing, fading, etc). To the best of our knowledge, WSNET is one of the simulators which have the most precise models of the physical layer. As we all know, it is always difficult to do experimentations. To implement the protocols on testbeds needs a lot of time (specially for debugging). It is also difficult to collect the running parameters of the sensor nodes in large-scale testbeds. In our case, reliability and delay are specially difficult to measure.
In this thesis, we propose protocols for real-time data gathering in WSNs. Other phases in a real-time application also have to be real-time, e.g. broadcast and aggregation. There are some cases where the sink node needs to broadcast some packets to the whole network. For example, the sink node wants to update the program or parameters of all the sensor nodes. Although broadcasting packets to the whole network is not a new research subject, to broadcast packets under ultra-low duty-cycle is a new research subject. There exist some broadcast protocols designed for ultra-low duty-cycle WSNs [Guo et al., 2009; Lai and Ravindran, 2010; Sun et al., 2009; Wang and Liu, 2009]. Nevertheless, they focus on the reliability and do not consider the real-time constraints. As a result, to broadcast messages to all the sensor nodes within a deadline for ultra-low duty-cycle WSNs is an interesting work. For example, how to design a broadcast protocol which can bound the broadcast delay, maximize the delivery ratio and minimize the number of broadcast packets?
After an event happens, many nodes may detect it and want to send packets. The generated packets are aggregated into an alarm packet and only one node sends it. For real-time appli- cations, it is necessary to have real-time aggregation algorithms because the interval from the event happens to the time when the packet is received by the sink must be bounded. So to de- sign a real-time aggregation protocol for WSNs is also an interesting work. For example, how to design an aggregation protocol which can bound the delay from the moment when the event happens to the moment when the packets from the nodes that detected the event are aggregated into one alarm?
According to the deadline of the network, we can divide the deadline into two sections, one for aggregation process and another for transmission process. Then the parameters of the protocols (e.g. WSEDR, RTCF or aggregation protocols) can be adjusted in order to satisfy the deadline and lifetime requirements. If both the aggregation and the transmission processes satisfy their own deadline, the deadline of the network can be satisfied. In this case, the param- eters depend on the requirements of the applications. Another point of view of this problem is
to consider the requirements of the applications and then give some guidelines for the deploy- ment.
Both the simulation results on a simulator and the experimental results on a real testbed can show the performances of a protocol, but they can not guarantee that the protocol can work as good as they want in all the different cases. By using the mathematic method, we have proven that the scheduling algorithm can optimize the delivery ratio. However, modeling our protocols with formal methods can validate the implementations of our algorithm. Moreover, in mathematical mode, we only consider the average link quality. With formal model, we can have preciser link quality models and then show precisely the performances of our protocols. So it is also very interesting to verify our protocol with formal validation method. For our protocols, the time constraints and probability should be modeled. Some formal verification tools can realize this, for example, PRISM [Hinton et al., 2006] which is a probabilistic model checking tool. The main difficulty is to take into account of the unreliable links. As the topology is dynamic, the number of states is increasing a lot. Therefore, it is impossible, in practice, to adopt this kind of tools directly to verify wireless protocols. Further works are needed to study how to use this kind of tools to verify wireless protocols.