sfi

NDP

ENTERPRISE IRELAND

NCNRC

Communications Network Research Institute

Congestion Control on WLAN Mesh Networks

Mustafa Ramadhan

Introduction

Wireless Mesh Networks (WMNs) are a type of radio based network which require minimal configuration and infrastructure. They usually consist of three types of nodes: client nodes, router nodes, and gateway nodes (see Figure 1). Clients can be either stationary or mobile, and can form a client mesh network among themselves and with routers. Wireless routers in a WMN relay packets originating from client nodes to gateway nodes which are connected to a wired network.

Figure 1
Figure 1: Wireless Mesh Network

Overview

Due to the shared nature of the wireless medium, a wireless link in a mesh network does not have a dedicated bandwidth since nodes in the vicinity may also contend for the same bandwidth. Therefore, an effective routing metric must be able to capture the contention for access to the medium between competing flows. We have modified the DSR protocol to make it better suited to the WMN environment. In this modification, a metric (AEF) that reflects the level of contention experienced locally at a node is incorporated into the route discovery mechanism. Since the nodes in the network contend for access to the wireless medium using the IEEE 802.11 DCF MAC mechanism, a high level of contention for access to the medium will result in a low availability of bandwidth at a node.

Problem Description

One of the main problems with WMNs based on the IEEE 802.11 standard is to realize bandwidth sharing among flows with different hop distances to the gateway node. Our objective is to develop distributed congestion control schemes for efficient bandwidth utilization. This can be done through implementing modifications to the Media Access Control (MAC) and Network layers.

We are investigating the problem of congestion control in WMNs. Congestion can occur whenever the incoming traffic to a node is greater than the outgoing traffic. The congestion problem can be tackled in a number of different ways: We can reduce the incoming traffic into a congested node by selecting an alternative route to send the data through the network; or alternatively we can attempt to manage the bandwidth by assigning more bandwidth to nodes with higher loads. This can be achieved by utilizing the QoS facilities available under the IEEE 802.11e/WMM standard.

Operation of the Modified DSR Path Selection Rule

In our work, the routing discovery mechanism of the DSR protocol has been modified by using the Access Efficiency Factor (AEF) as an alternative routing metric to the hop count. The operation of the modified DSR path selection rule has been modified by finding the path with the highest minimum AEF (max_min_AEF) value in order to avoid routing packets through areas of high congestion. The routing decision is made at the source node on the basis of a path selection rule which is in turn based upon a path selection metric (see Figure 2). The path selection metric implemented in this work is the AEF value measured locally at a node which is determined by the level of contention experienced by the node and by the load transmitted by the node (which is in turn determined by the routing decisions made by the path selection rule). The modified DSR path selection rule selects routes containing the bottleneck node that is least likely to become congested.

Figure 2
Figure 2: Congestion avoidance operation of the modified DSR path selection rule.

Performance Evaluation

The objective of our work is to utilize locally generated MAC layer information at the routing layer to improve the global performance of the network. Through computer simulation using the OPNET modeler it has been demonstrated that the route selection rules based upon the AEF metric exhibit better load distribution across the network nodes than the standard DSR path selection rule (see Figure 3). This figure shows the PDF of the load distribution for standard DSR routing algorithm and the modified DSR routing algorithm for 1000 random topologies with one gateway and 99 mesh nodes ( where each node has an average of 2 neighbours) randomly distributed across the network, packet rate is set to 5 packets per second, and packet size is set to 512 bytes.

Figure 3
Figure 3: Load distribution for the modified DSR against the standard DSR.

A better load distribution resulting from the modified DSR path selection rule improves the network performance, due primarily to avoiding the creation of heavily loaded nodes. This mechanism reduces the amount of dropped packets at the nodes (see Figure 4). To verify the throughput improvement of implementing the modified DSR routing algorithm over the standard DSR, the CDF of the packet loss ratio for 1000 network topologies where each node has an average of 2 neighbours are plotted, see Figure 4. For these topologies, the packet size is set to 512 bytes and the packet rate is set to 5 packets per second. Each topology comprises 99 nodes randomly distributed across the network and a single gateway at a fixed location.

Figure 4
Figure 4: CDF of the packet lost ratio for the modified DSR against the standard DSR.

From this figure, it be seen that for the standard DSR protocol 50% of the random topologies experience a packet lost ratio less than 22% compared with the new routing protocol where 50% of the random topologies experience a packet lost ratio less than 7.5%. It also can be seen that for a packet lost rate less than 10%. No topologies using the standard DSR routing algorithm achieve this performance compared to the new routing algorithm, where 82% of the random topologies experience a packet lost ratio less than 10%. Based on this result, it can be seen that, the new routing algorithm exhibits a significant reduction in the packet loss ratio compared to the standard DSR due to avoiding routing through highly congested nodes.

The performance of the newly introduced max_min_AEF metric has been evaluated against the Hc and the well known ETT metric. Three network scenarios where the number of neighbours each node has is 2, 4, and 6, have been examined. The nodes in the network have been divided into four sets of nodes one of which consist of 25 nodes with a transmission rate of 11 Mbps, the second set consists of 25 nodes with a transmission rate of 5.5 Mbps, the third set consists of 25 nodes with a transmission rate of 2 Mbps, and the fourth set consists of 24 nodes with a transmission rate of 1 Mbps. For each scenario the complementary cumulative distribution function (CCDF) of the throughput improvement and the delay increase for all network topologies examined have been calculated. The CCDF provides for a statistical characterisation of the improvement in the throughput and the increment in the delay produced by the new routing algorithm.

The CCDFs of the percentage throughput improvement and the average delay time increment were calculated for the new metric based upon the AEF against the standard DSR and the CCDFs of the percentage throughput improvement and the delay time increment were calculated for the DSR protocol based on the ETT against the standard DSR, see Figure 5.

Figure 5
Figure 5: CCDF of the percentage throughput improvement and average delay increment for DF = 2, 4, and 6 scenarios for AEF against ETT metric.

In all the simulation scenarios considered here it has been shown that the newly introduced max_min_AEF metric has shown a significant improvement in the global throughput of the network compared to the Hc and ETT metrics. However, this improvement in the throughput of the network is accompanied by an increase in the average delay time. This is because the modified routing algorithm avoids congested areas by routing packets away from the congestion, i.e. by taking longer transmission routes. Therefore the congestion avoidance strategy of the modified routing mechanism results in an increase in the average delay time of the network.