Predictive Congestion Avoidance in Wireless Mesh Network

1 downloads 8 Views 171KB Size Report
Aug 2, 2012 - Abstract—Congestion avoidance is a vital part in improving the Quality of service (QoS) in Wireless Mesh Network. (WMN). A preventive ...

2015 3rd International Conference on Future Internet of Things and Cloud

Predictive Congestion Avoidance in Wireless Mesh Network

Fawaz A. Khasawneh, Abderrahmane BenMimoune, Michel Kadoch

Mohammed A. Khasawneh Quality System Engineering Department Concordia University Montreal, Canada

Electrical Engineering Department ETS - University of Quebec Montreal, Canada

Abstract—Congestion avoidance is a vital part in improving the Quality of service (QoS) in Wireless Mesh Network (WMN). A preventive congestion avoidance algorithm is proposed in this paper which predicts the congestion before it really happens in the network by using different statistical analysis models which analyze historical traffic data in the network with some level of certainty to predict the future traffic data. Our proposed algorithm improves the Hybrid Wireless Mesh Protocol (HWMP) routing protocol used in IEEE 802.11s MAC layer standard. Based on the predicted link congestion, rerouting algorithm is implemented in order to assure the load balancing and to prevent congestion over WMN network. Simulation results show that our proposed algorithm outperforms other algorithms in the literature in terms of throughput, end-to-end delay, and fairness.

eliminates most of the limitations of previous standard and enables an infrastructure free, ad hoc, on demand and mobile wireless networking solution for remote and hard to reach areas. Hybrid Wireless Mesh Protocol is the built-in routing protocol for IEEE 802.11s standard that incorporates dynamic routing algorithms and effective in setting up reliable and flexible transmission paths. The main focus in designing this protocol was to make the network and devices technology independent such that devices that are not in each-other’s communication area can also communicate effectively [5]. This routing algorithm includes both reactive and proactive routing into the network [1]. The reactive elements uses radio metric for being compatible to reactive ad hoc on demand distance vector protocol. In its proactive form, the routing algorithm emerges routing information for reaching root mesh point proactively. This routing protocol incorporates features form Ad hoc On-demand Distance Vector (AODV) protocol. The main disadvantage of HWMP protocol is its inefficient response to congestion.

Keywords: HWMP; congestion control; wireless mesh network; QoS



The IEEE 802.11 standard is a wireless local area network protocol, which is largely dependent on wired infrastructure for data traffic transmission [5]. This standard was devised to extend wired network to wireless territory. There are a series of standards under IEEE 802.11, including a, b, g etc. which are related to different wireless medium. This standard specifies frequencies, mode of communication, modulation techniques and other specification for wireless mesh networks [2]. This standard entails two different types of networks including infrastructure mode where a definite access point is present and ad hoc mode, where no specific access point is present and two stations connect to each other on demand.

The congestion control mechanism is comprised of three operations which are monitoring and detection of congestion, signaling for congestion control and controlling of local rate. Different implemented congestion control protocols are designed to provide signals which contain information of the usage. Only the congestion control signaling part is defined by the IEEE 802.11s standard, whereas the rest operations of congestion control are not specified and left for vendor specification. The main operation of congestion control mechanism is initiated by a mesh station that is employing IEEE 802.11s congestion control signaling protocol, when it identifies any congestion. The stations immediately transmit a notification frame to its neighboring stations. The notification frame is known as Congestion Control Notification frame (CCNF). The main component of the CCNF is CNE (Congestion Notification elements), which can be one or more in number. The CNE contains two information-firstly, for every access category, it contains congestion notification duration timers, and secondly, it contains an address field for the destination mesh station. the second information is helpful in indicating the flow for

IEEE 802.11s standard is an essential modification to IEEE 802.11 standard that resolves wired infrastructure dependency problem of its predecessor. This standard works in link layer of OSI model and deals with physical address of the devices rather than IP address, which can be useful in layer-2, multi hop communication [2]. The standard proposes two different and novel path selection methods including RA-OLSR (Radio-Aware Optimized Link State Routing) and HWMP (Hybrid Wireless Mesh Protocol). This standard 978-1-4673-8103-1/15 $31.00 © 2015 IEEE DOI 10.1109/FiCloud.2015.69


this protocol incorporates Minimum Delay (MD) and Expected Transmission Count(ETX) metrics.

which local rate control should be implemented. Sometimes, the address field contains broadcast address, which compels the receiver of the CCNF to apply control mechanism to all flows or at least to flows that are intended for the CCNF sender station [4]. Thus a CCNF from a sender initiates local rate control mechanism in receiver stations based on the CCE content and congestion control is done.

Poor performance in multi-channel Wireless Mesh Networks resulting from congestion may be greatly reduced by Multi-Channel Adaptive Pacing (MCAP) scheme [7]. This scheme utilizes a cross-layer congestion control scheme. This scheme is potentially immune to congestion as it coordinates and transmits packets in-group during contentions and thereby improve multi-channel utilization.

However, congestion avoidance is better than congestion control because the latter is imposed when the network is already experiencing congestion. When the volume of queue at a specific node reaches to a pre-defined limit, it performs a buffer size check. If the queue size is near the buffer limit, the node starts broadcasting CCNF signals to its neighbors that use same path. The neighboring nodes then start searching alternative paths to send their packets to destination and avoid the path that is nearly congested as proposed in the Congestion Avoidance - HWMP (CAHWMP) [3]. Thus, the congestion is avoided. One of the added advantage of this mechanism is that it is not only helps avoiding congestion but also helps in load balancing.

Two proposed techniques are supposed to determine the protocol in [7]. Firstly, rate adaptive pacing where routers act like access controller to control the flow based on the bandwidth information and congestion measurement. These routers are known as ingress routers. On the other hand, another technique implemented at routers, which coordinates packet transmission locally to improve channel utilization by addressing the issues of exposed terminal and hidden terminal effects. These routers are known as congested routers. The protocol suggests that the congested router should transmit packets for a certain period taking into account the channel reuse constraints and then remain silent for a period, known as forbidden period, to allow transmitted packets travel further away. WCETT routing metric as incorporated with Adhoc On demand Distance Vector (AODV) protocol as multichannel routing protocol.

In this paper, an Enhanced Congestion Avoidance HWMP (ECA-HWMP) is proposed which predicts the congestion in each link in the network by applying two statistical analysis methods which are Croston and HoltWinter methods. The two methods rely on the historical data collected in advance to predict the future data. Since ECAHWMP predicts the congestion before it really happens in the network, the overhead produced by the congestion control signaling in HWMP protocol is reduced. However, the prediction process has some level of certainty (>95%) and in some cases (

Suggest Documents