Multihop Wireless Access Networks for Flood Mitigation Crowd-Sourcing Systems

Abstract

Natural disasters can be mitigated or even anticipated if we have appropriate means, in terms of communications and data sharing models, to collect relevant data in advance or during disaster occurrences, which can be used for supporting disaster prevention and recovery processes. This work proposes a framework that encourages people to collect and share data about disaster, especially flood in Ho Chi Minh City, via on-site established multihop wireless access networks configured by the sharing of internet connectivity in users’ mobile devices. For connectivity sharing, on-the-fly establishment of multihop wireless access network (OEMAN) scheme is thoroughly analyzed and improved to resolve its inherent issue on traffic load imbalance due to its tree-based structure. More specifically, we propose a linear program for overload-aware routing optimization considering wireless interference. Evaluations implemented in Matlab show that the overload-aware routing improves load balancing among available virtual access points in OEMAN. By avoiding nodes with heavy load in the network, our solution improves network throughput compared to overload-unaware routing protocols.

Appendix A Overload Detection in OEMAN

This section describes functionalities in OEMAN to handle overloaded nodes. Some of these functions have been modified to adapt them to our optimization model. Let $L_{i}$ denotes the load of node i $\in$ V. The load $L_i$ is defined as the total amount of traffic $u_{(i(e))}$ which is unicast on link e at node i minus the sending capacity $q_{i(e)}$ of node i on link e:

$$\begin{aligned} L_{i} = u_{(i(e))} - q_{i(e)} \end{aligned}$$

(8)

Overload is detected at a node i when $L_i > B_i$, where $B_i$ is the buffer size of node i. Overload detection is illustrated in Fig. 10. In this example, $PC_1$ is the VAP of $PC_3$ and $PC_4$, where PC is personal computer. $PC_1$ examines its load [Algorithm 1: line 7–9] by

$$\begin{aligned} L_{PC_1} = u_{PC_1(e(PC_1,AP))} - q_{PC_1(e(PC_1,AP))} \end{aligned}$$

(9)

where $q_{PC_1(e(PC_1,AP))}$ is the sending capacity of node $PC_1$ on the link $PC_1$ to AP and $u_{PC_1(e(PC_1,AP))}$ is the total amount of traffic from $PC_3$ and $PC_4$ which is unicast on link $e(PC_1,AP)$ at $PC_1$. $PC_1$ detects that it is overloaded ($L_{PC_1}$ ${>}$ ${B_{PC_1}}$) [Algorithm 1: line 7] since the traffic from two sub-trees t1 and t2 rooted by $PC_3$ and $PC_4$ is greater than the buffer capacity of node $PC_1$. In that case, $PC_1$ asks for help from other PCs for sharing load. Consequently, $PC_1$ will broadcast its overload messages to all PCs managed by it, namely $PC_3$ and $PC_4$. Immediately, $PC_4$ moves to another VAP ($PC_2$) [Algorithm 1: line 10–11] to transmit its traffic to AP. Examining wireless interference under the protocol model of interference [29] is also performed before doing a handover to $PC_2$. Consequently, load balancing among VAPs is achieved.

Fig. 10.

A node with heavy load in a tree-based network

Algorithm 1 describes the computing of load at a VAP and the execution of the handovers in the case of the overloaded VAP.

Algorithm 2 is the handover function which considers the conditions of doing a handover of any client node.

Algorithm 3 is the function returning the value of sending capacity of examining node.

Algorithm 4 is the function returning the value of a VAP which can be consider for serving handover of a client node.