multi-layer system that combines advantages of decentralized wireless networks and of multi-channel MAC usage. I. INTRODUCTION. Recent consumer market ...
High Capacity Wireless Mesh Networks: Principles, State of the Art, Challenges PREPARATION OF PAPERS FOR THE ARCHITECTURAL AND ENGINEERING SCIENTIFIC CONVENTION CCIA 2010 (TITLE) 1Main
Author (full name), 2 Second Author (full name)(do not fill this entry for the first version of the paper, which will be sent to the reviewers)
Affiliation, address, Institutional Affiliation, address
1 2 Christian K¨obel∗ , Walter Baluja Garc´ıa∗ , Joachim Habermann† , Eric S´anchez† and Emre Kocak† e-mail:
∗ Department Abstract— The abstract must appear on thePolytechnic first page of the paper. These instructions giveJos´ youeguidelines for preparing papers for the of Telecommunications, University ISPJAE Antonio Echeverr´ ıa, Havana
Cuba
Architectural and Engineering Scientific Convention CCIA´2008. The proposed format is very similar to that used for IEEE journals. Email: {Kobel, Use this document as a template if you are using MicrosoftWalter}@Tesla.cujae.edu.cu Word 6.0 or later. Otherwise, use this document as an instruction set. As the † Department document will be included in CD-ROM, the maximum number of pages is 10 and there is a limit for the & memory space that occupies the for Information Technology, Electrical Engineering Mechatronics file that contains the final document (PDF or PostScript format). This file should occupy a maximum of 4MB. The abstract should University of Applied Sciences FH Giessen-Friedberg, D-61169 Friedberg Germany contain the most important aspects of the paper, including procedures, results and conclusions of the work, in less than 150 words. Define all symbols used in the abstract. Do not cite references in the abstract.
Email: {Joachim.Habermann, Eric.Sanchez.Salas, Emre.Kocak}@IEM.FH-Friedberg.de
Key Words—About five key words or phrases in alphabetical order, separated by commas.
GUIDELINES FOR PREPARING ARTICLES FOR CUBAN JOURNAL OF ENGINEERING (Title in English)
Abstract—802.11 hardware has ultimately found its way to the signals for example. So what has been achieved for wireless ABSTRACT. give in youhome guidelines for preparing as proposal to Editorial of the Cuban for quality of Service (QoS) electronic consumer market.These Butinstructions especially networks it papers connectivity has to Board be achieved of Engineering clashes with high Journal quality demands of rich multimedia streaming on future wireless links. Therefore, basically the aspects of applications, that WirelessLAN can only partly cope with. This paper must have a maximum of 10, and not bandwidth less than 6, capacity have network robustness, stability and work describes basic features and fundamental flaws of 802.11, The pages, including figures, tables and references. Try to be to be improved. 802.11 is still based on the fundamental particularly with regard to multimedia streaming and Quality-ofconcise, and whenever possible, use short sentences to express I. INTRODUCTION Service (QoS) requirements. To overcome current problems, the design issue that its spectrum is not fully exploited in most your ideas. Do not include too detailed figures and tables that The introduction of the work should contemplate the be difficult to readfor and understand. presented work points towards multi-interface extensions setups, although example IEEE 802.11a, which is partly essential aspects, possible including the bibliographical revision and the might of 802.11-based networks, state of the art.in combination with decentralized deployed in our research context, would provide up to 12 Wireless Mesh Network (WMN)is structures. clearWord goalversions is to Title of the Paper and Authors This document a template for The Microsoft Write the title of the paper, for the names of the authors, their Multi-channel / orthogonal channels simultaneous usage. provide an overview all relevant 802.11 and WMN protocols institutional 6.0 oroflater. affiliations and e-mails on the first page. Provide a multi-interface WLAN networks offer a viable solution for with reference to important research works and to indicate title in English. The title must be centered, in 12 pt Times New crucial success factors and challenges of a future QoS-ready, Roman these bold-type problems and some aspects Leave of their uppercase letters, key single-spacing. a 12 exploitation will II. MATERIALS AND METHODS multi-layer system that combines advantages of decentralized pt blank spacebebetween the titlein andthis the authors’ therefore presented work. names. The Preparation and extension of the text. title must not be longer than 80 characters, including spaces. wireless networks and of multi-channel MAC usage. Furthermore, this work highlights multi-channel approaches in
Write your paper directly in this format. Use 10 pt Times New Roman fonts. Leave an extra blank space at the beginning of I NTRODUCTION every I. paragraph. Use an impersonal style. Abbreviations must be omitted and acronyms should be clearly defined. Avoid Recent consumer market tendencies reveal that IEEE Wirewriting lists with numbers or symbols. The paper must begin theby “Introduction” on the firstfamily page, then two blank lessLAN 802.11 with is the far most popular of wireless spaces must be left at the end before “Key Words”
standards for consumer networking. Decreasing manufacturing Extension. sizes and price development allow most mobile devices like smart phones, tablet PCs, netbooks and cell phones to include WLAN hardware. This further led to the trend that manufacturers of bigger consumer equipment like Internet-ready televisions now also implement 802.11 and thus bring forward the linkage of several media platforms (on a consumer application level). All things considered global consumer industry has made great experiences with WLAN. Valuable and diverse contributions by researchers dealing with 802.11 networks reflect the general acceptance. But the increased usage of wireless connections has made wireless networks, especially home networks, become more heterogeneous concerning their client structure. At the same time, new media applications raised the requirements on wireless networks and users expect the same quality of experience (QoE) from web based service platforms (e.g. YouTube) as it is as known from broadcast TV
The title of these instructions is 100 characters long. There must be a maximum of four authors and who must be proprely(WMN). In most combination with Wireless Mesh Networks identified with WMNs superscripts, their institutional affiliations, use cases, are based on WLAN access technology. addresses and e-mails. Write the authors’ full names, WMN models decentralized network architecture, beginning by the firstdescribe and middle anames, followed by the last name(s), centered, under enabled the title. Use Times New which is basically by 10thept ad-hoc mode. This mode Roman, bold-type, upper- and lower-case fonts. Leave a 10 pt is provided by the allauthors’ 802.11 compatible interfaces, besides the blank space between names and their affiliations. Write the infrastructure affiliations and their using mode. upper- A common inwiderdown used (oraddresses, managed)
frastructure scenario includes an access point (AP) that forms a multipoint-to-point Basic Service Set (BSS) cell with its clients. Routing is simple in this case, but a main disadvantage of this setup emerges when the AP breaks down, since all connected clients lose connectivity at once. To enable roaming between APs for a client, it is possible to combine several BSS into a Extended Service Set (ESS), using the 802.11 Wireless Distribution System (WDS). Practical experience reveals that roaming functionality is mostly limited when using different vendor hardware, due to compatibility issues. Hand-off connections between two APs also offer poor QoS performance, due to a time-consuming channel re-allocation scheme [1] which forces the client to scan all available channels in order to locate the new AP. Because of these issues most infrastructure setups, examined in a larger scale, form only small, isolated
islands of wireless coverage with a wired backbone connecting the APs, which further act as Internet gateways for their local wireless clients. Nowadays wireless infrastructure setups may still be more common, but recent hardware development favors the shortfall of central routing elements such as APs, since nodes provide more CPU resources and are able to compute complex routing decisions and forward packets in mesh networks on their own. To decrease the dependency on wired backbones and to enable a more scalable collective network, with enhanced all-wireless coverage and connectivity over a larger area, a WMN is the most suitable approach [2]. The nodes of a WMN form a self-configuring, self-healing and self-maintaining mesh cloud. Due to the open structure, new nodes can be flexibly integrated. Mesh architecture sustains signal strength by breaking long routing distances between two nodes (that are not connected through a direct physical link) into a series of shorter hops, following the multi hop principle. Intermediate stations work as packet forwarding nodes in a cooperative manner here. Since no central routing instance is needed, a WMN is perfectly scalable when it comes to extending the network, e.g. in order to eliminate radio frequency dead spots that a common infrastructure access point could not reach. This advantage is in particular valuable in wireless home networks, where often a single AP is not sufficient to connect all clients over several floors. Home networks with the purpose to essentially transport real-time media and less data traffic (e.g. a file download) are a key motivation scenario of this work. Due to the wireless nature of links in a mesh cloud the topology might change infrequently; new links come up, existing links break down or suffer from negative radio propagation effects, even if the nodes are immobile. Research communities established efficient solutions to conquer these conditions, in form of open-source mesh routing protocols like OLSR or AODV (see ch.IV) which aim to compensate unstable or lossy links with alternative routing paths. IEEE standards committee also pays tribute to this development by implementing mesh architecture into the new wireless 802.11s sub standard [3]. But still, basic connectivity may be mostly granted in WMNs, but due to the mentioned diversity of the consumed media contents also the traffic in WMNs becomes less predictable and heterogeneous. Thus future efforts must concentrate on increasing QoS performance. The covered protocols in this article are presented in consideration of an underlying multilayer philosophy and the upcoming design challenges that come with cross-layer driven, QoS-oriented multi-channel systems. The rest of the work is structured as followed: Chapter II recesses the topic of wireless mesh network architectures and its most common applications. Relevant routing mechanisms are highlighted in chapter IV. Chapter III provides essential and state-of-the-art knowledge about underlying PHY and LLC protocols described by the 802.11 standard. Chapter V deals with Quality-of-Service demands generated by higher network layers. Finally, chapter VI shows off challenges and chances of possible multi-channel solutions for WMNs. In chapter VII representative 802.11 measurements are presented
Ethernet / External Host Network
Mesh Portal / Gateway
Mesh Portal / Gateway
Mesh Client Mesh Router
Mesh Client 802.11 Link
Fig. 1.
Non-Hierarchical Mesh Network
Ethernet / External Host Network
Mesh AP
Mesh AP Local Infrastructure Client
Mesh AP with Gateway
Mesh AP
Mesh AP with Gateway
Mesh AP
Local Infrastructure Client
Fig. 2.
Hierarchical Mesh Network
and chapter VIII concludes the presented approaches. References to related works are included throughout the article. II. A S URVEY ON M ESH A RCHITECTURES Wireless Mesh Networks mostly consist of mesh nodes, mesh routers / mesh access points and gateway nodes. Figure 1 depicts a non-hierarchical scenario in which all nodes (regardless of the type of device) fulfill packet-forwarding functionality and multi-hop connectivity is established across the entire network. This scenario is mostly applicable for wireless home networks. WMNs can include network hierarchies as well (simplified in figure 2) where users are locally connected to mesh APs, which form the actual mesh cloud and provide Internet or external host network access. Furthermore hybrid mesh networks may combine both hierarchical and nonhierarchical structures. Fields of application are diverse and may reach from semi- and full-mobility scenarios, metropolitan areas, corporate or academic setups over to vehicular networks. The scale of a WMN then varies with number of hops and hop-to-hop distances. First WMN setups in Germany, like the Freifunk urban community networks [5], started off with combining private access points to an open mesh backbone, to share unevenly distributed Internet access points within the community. The effectiveness of such quickly adapting deployments in combination with its cost efficiency (no new hardware is needed, all is handled by open source routing protocols) made mesh network architecture attractive for developing and threshold countries as well. Countless initiatives and projects, most notably in Africa [6], rely on mesh based wireless backhauls to overcome for example inaccessible geographical areas (e.g. [7]). Most WMNs have in common that traffic is not equally distributed among routes. Especially in the private sector (e.g. home network) users frequently access external Internet resources and thus routes to the gateway node are more often
penetrated than pure node-to-node connections [4], which provokes bottlenecks near the Internet gateway. If routing measures cannot compensate such limitations, alternative solutions have to be found in the underlying 802.11 access technology. WMN architecture has other restrictions that are not directly relevant for QoS performance. Optimized security protocols are not sufficiently refined, due to the open character of a decentralized structure and the missing central instance to control security breaches, perform intrusion detection and enable data authenticity. Also the power consumption per node is hardly controllable, in contrast to an infrastructure setup where the access point automatically adjusts tx power levels. A cross-layer metric (see chapter IV produces relief in this case. III. 802.11 - OVERVIEW AND R ELEVANT C OMPONENTS The IEEE 802.11 standard covers the Medium Access Control (MAC) sublayer and the physical layer of the OSI reference model. But an effective QoS-sensitive transmission chain requires the revision of QoS-related parameters (e.g. CPU power, buffer sizes, encryption, etc.) on almost every OSI layer. In this chain, 802.11 sub-protocols, especially in MAC layer, are at risk to become a bottleneck [9]. This chapter now covers the threat potential of known 802.11 issues in reference to QoS constraints, in order to efficiently apply mesh structures in home networks. A. 802.11 Physical Layer Schemes A high dependency on whether the line-of-sight between sender and receiver is free or obstructed limits common 2.4 GHz band 802.11b/g links to a maximum connection range of approx. 46 meters indoors and approx. 92 meters outdoors. Larger distances result in unacceptably high packet loss rates and packet jitter. Using 5 GHz band 802.11a even cuts these values in thirds, because 802.11a radio hardware employs a higher signaling frequency, which makes transmitted waves more susceptible to obstructions. Unfortunately, the actual interference range is a multiple of these values, in the worst case up to a factor of 10. Due to the popularity of WLAN, most indoor scenarios are already indirectly affected by neighboring access points for example, which makes interference on the physical medium barely controllable, especially in the mostly crowded 2.4 GHz band. Besides adjacent- and equal- channel interference even other product groups like e.g. microwave-ovens might lower the signal-to-noise ratio. To conquer interference problems, 802.11 physical layer natively provides frequency diversity: Depending on the region 802.11b/g offers between 2 and 3 and 802.11a up to 12 nonoverlapping channels [8]. This generally favors the deployment of mesh nodes equipped with multiple wireless network interface cards (WNIC), but current WLAN stations mostly deploy only a single interface which can only communicate on one channel at a time. The first standardized attempt (in consumer hardware) to increase spectrum utilization within a single WNIC is described by 802.11n [9]. Besides MIMO antenna features 802.11n allows to bond two 20 MHz channels
and thus to increase the bandwidth on the PHY layer. Still, applying 802.11n physical bonding in 2.4 GHz band would consume up to two-thirds of the available spectrum and might hence increase interference levels. This issue leads to the later proposed dynamic bundling approach which envisions MAC coordinated channel bundling. Applying several WNICs in one station or handheld is a promising approach to increase capacity of wireless links. But due to physical limitations, PHY layer does not always match the requirement that “using n interfaces equals n-fold performance” [10] and so only offers a limited modularity for WNIC combinations. Per node performance (e.g. regarding throughput) is also limited by CPU capacity, which becomes more occupied with every new WNIC [11]. Alignment of wireless devices and distances between antennas might also jeopardize performance because of disadvantageous radio propagation [12]. In reality propagation effects also might negatively affect orthogonal channels, which in theory are supposed to be interference free [13]. Despite of these limitations, capacity gain is always superior to all single-interface setups. B. 802.11 MAC Layer Schemes Besides physical interference, also the chance of a high amount of nodes sharing the same coverage area is responsible for the inherent unreliability of wireless systems. The more nodes are active, the higher is the packet congestion: Although 802.11 PHY layer is designed for multi-channel purposes, the medium access control (MAC) scheme implies a shared channel for all stations, using CSMA/CA (Carrier Sense Multiple Access with Collision Avoidance). This scheme was basically designed for point-to-multipoint communication (client to AP) and has its roots in Ethernet-based networks (CSMA/CD). Adhoc networks now have fundamentally different requirements though, since every node is supposed to have multipoint-tomultipoint connections to different next hops. CSMA/CA is realized by the Distributed Control Function (DCF), situated in the lower sub-layer of the data link layer. The DCF protocol coordinates the medium access of several WLAN stations with use of the MAC overhead packet types Request-toSend (RTS) and Clear-to-Send (CTS) (and the corresponding acknowledges). DCF not only aims to decrease packet contention on the channel, but also prevents hidden node problems [8]. This issue frequently occurs if two nodes, that are invisible (hidden) to each other, attempt to communicate with a common neighbor on the same channel. They both sense that the carrier is free and start to send simultaneously, causing packet contention. CSMA/CA was originally designed for solely best effort transmission, which means that all packets are treated equally, not considering prioritization of different packet classes. Applying CSMA/CA with a high network load implies that a packet might be delayed for an uncertain amount of time before it can be transmitted. Data packets can handle larger delays than voice and video packets, which allow them to be buffered and transmitted in a best effort manner. Coherent real-time multimedia streams on the other hand are more sensitive to delay and jitter of single interrelated
packets. In practice this means that sometimes bursty best effort traffic may choke out a long-running, prioritizationworthy video stream, thereby destroying the user’s experience. The lack of QoS support in 802.11 led to the first standard amendment 802.11e [14], that takes different traffic priorities into account. In 802.11e, standard DCF (with its strong best effort character) is replaced with the Hybrid Coordination Function (HCF), which increases the transmit probability for QoS-sensitive packets. 802.11e performance strongly depends on packet inspection methods in the MAC layer in order to detect protection-worthy traffic, or on the corresponding QoS signaling in IP headers, performed by upper layers (e.g. DiffServ). This denoted dependency underlines the advances of a cross-layer QoS system [15]. Because of their poor scalability, random access protocols such as CSMA/CA are not an efficient solution for WMNs. Standard 802.11 MAC is not designed for multi-hop communication, which creates elementary problems in intermediate nodes which forward packets between source and destination. These packets have to be replicated multiple times through multiple nodes, but still belong to the same end-to-end flow. The 802.11 standard does not consider this and forces every to-be-sent packet to compete for the medium anew, although it would be sensible to coordinate common time slots for coherent forwarding packets on all involved intermediate nodes on the multi-hop route. In affect, the available throughput for each node is not only limited by the raw channel capacity, but also by the forwarding loads imposed by other nodes, since only one node can access the medium at a time. This results in an inefficient store-wait-and-forward process along the route. To reduce overhead time wasted for channel access negotiation, Bononi et. al. [15] described a basic and a fast-forward negotiation mode. The former node equals standard DCF procedure, the latter mode extends RTS / CTS messages by a flow identifier and by a variable reservation period for each forwarded stream, facilitating multi-hop transmissions. These schemes can be further optimized by including a distributed multi-interface usage, for example when a set of interfaces A is responsible for the actual data transmission, while another set B coordinates the synchronized medium access along the route. Such enhanced functionality, when packet forwarding indicated by the network layer is supported by link layer, further argues for a cross-layer approach. Therefore an according 802.11 amendment could effectively reduce end-to-end delay, which favors QoS streaming. IEEE standard committee recognized these needs and has formed the Robust Audio/Video Streaming Task Group, who defined general QoS requirements on WLAN in the 802.11aa standard amendment [16]. 802.11aa mainly depicts enhanced signaling between APs/stations and robust multi-cast, but also relies on packet inspection methods. A general overview of related 802.11 amendments is given in [17]. IV. M ESH ROUTING P ROTOCOLS Routing in a decentralized wireless mesh network is more complex than in hierarchical infrastructure networks, because
there is no centralized hardware present to coordinate routing tables. To enable multi-hop functionality, a joint, decentralized routing engine is necessary. In order to allow every node to establish a correct routing table that covers routes to all destinations in the mesh cloud, certain information is interchanged between nodes. Depending on the type of that information, currently used protocols can be roughly categorized into Link State and Distance Vector routing protocols. Further subcategories include proactive protocols, which establish routes independent of their actual usage, and reactive protocols, which calculate routes on demand. Link state routing protocols, like its most popular representative OLSR [18], share a minimum of link state information between nodes, but in a network-wide manner. Every node then individually calculates its routing table based on these link state messages. In contrast, distance vector protocols like AODV [19] have nodes share their entire routing table, but only within their 1-hop neighborhood. Both protocol groups use different algorithms to discover newly acquired routes: Dijkstras algorithm is frequently used in link state protocols and Bellman-Ford algorithm in distance vector protocols. Regarding multi-interface networks with QoS appliance, distance vector protocols are suitable in principle [20], because they already share a lot of neighbor information locally, which could be extended with channel allocation information of the involved interfaces. Also, there is no need to distribute channel information network-wide, because of the limited coverage area. On the other hand DV protocols are less scalable and produce high amounts of overhead, since comprehensive routing information is exchanged. In addition, DV protocols do no converge fast enough, due to the unsynchronized and unacknowledged way that distance vector information is exchanged. A fast convergence might be a critical factor for a home network, since users expect multimedia devices to be quickly integrated into the network. Link state protocols like OLSR offer contrary and thus positive characteristics: Low network overhead (since exchanged information only contains single link characteristics), good scalability and fast convergence. Also, nodes need to process advanced information for calculating routing tables, which is compensated by the fact that consumer multimedia devices in home networks have often powerful CPU and memory resources. Therefore, in our simulations and testbed OLSR is used as the mesh routing protocol of choice. Routing metrics mark a strong performance criteria for the efficiency of the route finding algorithm because they determine the way in how a next hop link is evaluated in terms of quality. Current routing protocol projects like OLSR implement simple metrics like the hop count between source and destination or Expected Transmission Count (ETX) [18] in order to maintain the systems portability and independence of the access technology (Ethernet, WLAN, WiMAX, PowerLine, etc.). Research works on the other hand show that cross-layer approaches may heavily increase performance in mesh routing [20]. Including medium-specific characteristics like e.g. actual throughput, packet delay or packet delivery ratio values, directly obtained from driver API, allows a precise evaluation
of the link quality, but at the expense of compatibility (the system then always depends on a special operating system and WLAN driver). For a better multimedia performance in home networks, standard metrics should be replaced with residual bandwidth capacity and round-trip-time, which plays to the requirements of a video/audio stream. For a best effort traffic transmission high speed and low delay are not necessarily mandatory to carry out its duty, but rather a high packet delivery rate. To enhance this multi-layer metric concept, in a next step routing metrics should adapt to changing network conditions [21]. Given the assumption that traffic is spread heterogeneously, the cumulative equation that describes the used routing metric in total should include several sub-metrics and weight them with a factor that depends on dynamic zones in the mesh network. In zones with low QoS-related traffic activity and low packet collision/contention probability, it is better to traverse packets as fast as possible through this area, rather than observing the general bandwidth utilization in this zone for example. In this case packet delay and hop count dominate (represented by a higher weight factor). In zones with high network loads, that might threaten QoS performance on busy links, it is sensible to consider the residual bandwidth capacity instead. To realize a proactive indication whether a part of a route is about to carry real-time traffic or not, a source node could broadcast network-wide QoS-reservation messages for a planned endto-end route, before starting the actual stream. K¨ohnen et. al. [22] already implemented such functionalities in Ethernet networks. The reservations are included in the cumulative metric equation as weight factors. V. T RANSPORT L AYER AND B EYOND The growing presence and diversity of multimedia services results in higher amounts of payload, which simultaneously contains more QoS-related material. A major contribution to this trend is the broad acceptance of new standardized video resolutions like 720p/1080i/1080p in the home consumer market. Also, the interest in 3D video will further raise the bar for new levels of QoS-demands, as the user expects a constantly high quality of experience that matches the quality of the consumed high definition material. The streaming requires a high network bandwidth. A single 720p audio/video stream at 60 frames/s using 10 b/color for instance consumes about 1.4 Mbits/s [16]. Modern digital file formats like .mkv enable a high compression rate but compression cannot entirely compensate low bandwidth bottlenecks. Additionally, advanced scenarios in a home network might require multicasting of local video/audio sources, which is contrary to common pointto-point data communication and may further provoke bottlenecks (see figure 3, A)). To implement Quality-of-Service in a WMN, the correct identification and classification of realtime packets is mandatory, but also hard to accomplish, due to the convergence of Media over IP traffic (e.g. YouTube, Hulu and Telekom Entertain), best effort traffic (e.g. HTTP protocol) and linear DVB broadcasts. Traffic can be identified in several ways and the identification process should never
A)
B) Multicast Receiver
Multicast Receiver 2 IF
2 IF
2 IF 1 Shared Channel (1 WNIC / node) Vulnerable Link When Streaming Multicast Source
Internet Gateway
Fig. 3.
2
4 IF
3 Different physical channels 1-3 Multi-IF Nodes
Multicast Source
Internet Gateway
Single-Interface / Multi-Interface Home Topology
slow down the actual packet processing of the node. Most of the current setups require a deep packet inspection with a thorough payload analysis [22] to determine the real nature of a packet’s content, which is generally not a problem with common home consumer devices. After a successful identification, transport protocols like the Resource Reservation Protocol (RSVP) can reserve bandwidth for certain flows. RSVP requires that every client supports the protocol and is therefore mainly applied in homogeneous cable backbones and less in wireless mesh networks. Besides bandwidth reservation and packet prioritization (e.g. 802.11e) the consideration of a multi-interface environment marks a third viable alternative to protect traffic. The benefits are regarded in the following chapters. VI. C HALLENGES OF M ULTI -I NTERFACE E NVIRONMENTS This chapter attends to initial problems of multi-interface / multi-channel (MIMC) networks. The provided aspects are related to our work in [23], where technical details of a corresponding novel layer 2.5 approach are further described. To improve network capacity, multi-interface nodes are a viable solution. In their case it is sensible to have a single virtual MAC interface (which is presented as such to upper layers) combine and manage the connected real interfaces. This architecture provides a good tradeoff between portability and channel diversity exploitation, since the original 802.11 standard needs no further modifications and future PHY modulation schemes such as 802.11ac [17] can be easily adopted. Xu et. al. [24] examined differences between a clean multichannel multi-interface network, which implies assigning as much different channels as possible to different neighbor-toneighbor links, and sole channel bonding, which means aggregating bandwidth capacities of the used interfaces between two neighboring nodes. The group states that under high traffic load conditions, with a high degree of packet collisions that deteriorate the throughput and delay performances, MIMC greatly reduces collisions and may improve both throughput and delay since traffic can be equally distributed on more channels. Channel bonding only slightly decreases packet delays in high-traffic conditions. On the other hand, when traffic is low, the bonding of two interfaces achieves low delay values on a link and therefore favors QoS streams. Bonding
has a better leverage effect when applied to concrete links that form parts of routes carrying QoS-packets between two nodes. MIMC is rather suitable to balance and absorb high packet quantities in the mesh network. The two concepts are regarded and shall be combined in the following subsections.
Source Node 3
Source Node 1 2
3
3
Neighboring Node 1
Forwarding Node
3 Neighboring Node 2
3
Neighboring Node 1
Forwarding Node 5 3 4
3 Neighboring Node 2
A. Channel Distribution in MIMC Environments Channel distribution for multi-interface nodes must aim to at least equally distribute available channels in a hopby-hop fashion, in order to decrease per channel congestion and mutual interference levels. Also, in this way the network topology might be actually shaped through (physical) route separations by frequencies; for example to separate three different streaming/multicast single hop links, as depicted in the in-house scenario B in figure 3. But first of all channel distribution has to be applied in a way that a logical connectivity (appearance in the routing table, despite of the used channel) between a node and every neighbor within its WLAN coverage range is guaranteed. If nodes transmit on different channels without further knowledge of the channels their neighbors are reachable on, they become invisible to each other. In the literature this problem is reffered to as deafness, which further leads to common hidden terminal issues on multiple hops [25]. Therefore, to provide a common basis medium, a dedicated control channel [23] is needed. The scope of the control channel (CC) ranges from sole transmission of signaling packets generated by the MAC (RTS, CTS, ACK) and network layer (e.g. OLSR HELLO and Topology Control overhead traffic) to hybrid solutions where overhead traffic is mixed with best effort traffic. The purpose of the CC should consider a good trade-off between simplicity and utilization of CC bandwidth capacity. In any case CC must be accessible by all stations and must carry information about a node’s general access to a set of channels and which of them are currently assigned to / used by its interfaces. To somehow consider the limited coverage area, for the sake of simplicity a node distributes this information only to its 1-hop and 2-hop neighborhood, to not provoke excessive overhead by network-wide flooding of its channel signaling information. The information itself can be distributed through an arbitrary message type or, more economically, embedded in existing signaling messages. Another deciding design attribute of a multi-channel / multiinterface network is the question whether channel allocation is performed pro- or reactively. Bononi et. al. [15] deal with the question if channel assignment should follow after the actual route finding mechanism, or vice versa. Wu et. al. [26] implement channel allocation in a reactive manner by combining RTS/CTS management with the identification and distribution of unused channels within the 1-hop neighborhood, all based on a common CC. A proactive solution on the other hand might be more suitable, since a basic connectivity between nodes has to be established anyway. Assigning channels in reaction to a transmit request by upper layers only consumes additional channel (re-)allocation time; a critical factor for delay-sensitive real-time traffic. Additionally, interference is
Internet Gateway Scenario A: Single-Interface Nodes Used for both best effort traffic AND video stream from source to gateway
Fig. 4.
Internet Gateway Scenario B: 4 Interfaces per Node Used for best effort traffic Used for video stream from source to gateway
Separate Resources for Gateway Routes and QoS Traffic
prevented a priori when channels are assigned proactively before upcoming flows. The dynamic adaption to actual traffic loads and QoS-demands in a node, instead of simply assuring a uniform distribution of channels, make assignment mechanisms even more effective. To classify next-hop links and thus the majority of traffic they are bearing, we distinguish between best effort links (low priority), links that carry traffic towards or coming from Internet gateways and links that carry QoS traffic [23] (both high priority). Identification of the type of traffic may be provided by upper layers (deep packet inspection methods). Gateway nodes are identified through the routing table. As mentioned before, gateway routes generally carry the majority of traffic in WMNs and should receive more resources to prevent bottlenecks in the mesh region near the actual gateway node. The distributable channel resources depend on the amount of available interfaces per node. In a first step, QoS- and gateway routes are favored with separate interfaces, while best effort links may communicate on a shared channel, applying CSMA/CA. Figure 4 depicts a related scenario. In this case a video stream that takes a gateway route receives 4 different orthogonal channels, which significantly increases bandwidth, lowers delay and prevents negative interference effects for this stream. Based on this example, resources in MIMC environments are either represented by separate channels per neighbors, or in the advanced case, by multiple separate channels per neighbor, which lead to the application of channel bundles (see next section). B. Channel Bundling in MIMC Environments The described channel distribution scheme is handled proactively to enable a basic connectivity from the start. Scenario B in figure 4 offers abundant interface/channel capacities for the QoS-stream, which allows a bundling of the involved interfaces. Bundles can be dynamically generated respectively degraded, depending on the time-variable amount of 1-hop neighbors (assuming that links suddenly break down or come up). In this particular case, each link aggregates capacities of two interfaces. Bundles are formed reactively (for example directly after the identification of a concrete QoS-stream), when extra capacity is needed and if the involved nodes can provide sufficient interface resources. Channel Bundling requires load
VII. M EASUREMENTS We present a selection of measurements and simulations that reflect some key performance aspects of 802.11. A first simulation with the Network Simulator NS-2 depicts the prospective throughput (cf. figure 5), packet delay (cf. figure 6) and packet drop rate (cf. figure 7) when routing over multiple hops on a shared 802.11g channel (54Mbits/s raw bitrate). Each node is equipped with one WNIC. The presented simulation contains 8 nodes, arranged in a chain topology with a distance of a 100m between each node. OLSR is used as a mesh routing protocol. A free space propagation model is applied and the high distance assures that each node can only communicate with its 1-hop neighbors. The source node is placed at one extreme of the chain and initializes a 1024 kbytes/s CBR UDP stream with a packet size of 512 bytes and a duration of 200s to the destination node, which is placed at the other extreme and is thus 7 hops away. Multi-hop routing significantly increases the topological coverage of a node, but our results in a single-interface standard WLAN setup also show that throughput decreases, respectively that packet delay and packet drop rate increase with every new hop, becoming hardly satisfying for QoS-
Throughput (kbytes/s)
800 600 400 200 1
Packet Delay (ms)
Fig. 5.
2
3
4 5 Hop Count
6
7
End-to-End Throughput vs. Hop Count
100 50 0 1
Fig. 6.
2
3
4 5 Hop Count
6
7
End-to-End Packet Delay vs. Hop Count
streaming after three wireless hops (compare also [27]). This linear decrease of performance over 7 hops is notable, because it is contrary to the fact that the overall amount of nodes within a shared coverage area is always 2 or 3. The effect occurs due to the two conditions that the packet queue size per interface is limited and that each intermediate node cannot directly retransmit the packets with the same bandwidth they arrive with, since a 802.11 channel is not full-duplex and therefore the interface has to receive and send in a chronological manner. These two loss factors are carried on with every hop and cause a constant performance decrease. Multiple WNICs per node might dampen this effect due to separate TX and RX buffers. In this way the limited single-channel resources are spatially disseminated on other frequencies, optimizing channel utilization. Packet Drop Rate (%)
balancing. To double bandwidth capacity, both channels can be loaded simultaneously with packets. In [23] we propose to schedule more data packets to be transmitted via the channel with the best quality and most residual bandwidth, to fulfill QoS-requirements. Another concept sets the focus on stability. One channel might act as a fallback option in case the main channel (that carries all packets) fails, or if the packet error rate on the main channel exceeds a threshold. Alternatively, redundancy (multiple transmission of the same packet) might increase stability as well. On the arrival side packet reordering has to be performed, depending on the chosen load balancing scheme. Packet reordering is a critical performance factor as well since wireless links are less predictable and packets may not always arrive in the correct order, due to different delay values of the interfaces. In current research works channel bundling options are still poorly considered, mostly because it is supposed that nodes may use all interfaces to serve different neighbors first. Therefore additional WNICs for bundling are rarely available. Still, channel bundling is the future concept to increase network capacity in 802.11. Tradeoffs between MIMC and bundling have to be defined here to enable efficient hybrid scenarios like in figure 4, where all low-priority links can be served on a shared channel. Channel bundling decreases packet delay times and for this reason it is especially interesting for supporting media streams. Since the overall transmission bandwidth of a link that contains bundled interfaces is enlarged, the transmission rate increases, which decreases packet transmission time, assuming a constant packet size. Upper layer sizes must also be adapted then to aggregated WNIC receiver queues in the link layer, to not slow down QoS flow processing.
80 60 40 20 0 1
Fig. 7.
2
3
4 5 Hop Count
6
End-to-End Packet Drop Rate vs. Hop Count
7
·106
7
7
6
6 RTS / CTS Rate (bits/s)
RTS / CTS Rate (bits/s)
·106
5 4 3 2
5 4 3 2
1
1
0
0 500 1000 1500 UDP Datagram Size (Bytes)
500 1000 1500 UDP Datagram Size (Bytes) 0.5 Mbits/s Fig. 8.
1 Mbits/s
5 Mbits/s
20 Mbit/s
RTS/CTS Rate vs. Packet Size with HC=1
A real-life testbed is used for the following measurements. It consists of 2 nodes (cf. results in figure 8), respectively 3 nodes (cf. results in figure 9) aligned in a chain with a distance of 15m between each neighbor. WLAN channel 1 with 2.412 GHz is used and a fourth node in Monitor Mode measures the rate of RTS/CTS control packets generated by the 802.11 Distributed Control Function. This rate maps the number of to-be-transmitted packets that have to compete for their transmission, or in other words, the degree of packet congestion on a shared medium. A source node starts a UDP stream that varies depending on a combination of 3 different UDP packet sizes and 4 different constant bitrates. In the first setup (figure 8) the destination node is its direct neighbor, while 1 intermediate node is placed in between in the second setup (figure 9). As expected, RTS/CTS rate increases with a higher CBR, since more packets are transmitted with a constant packet size. At the same time, with a constant UDP bitrate, the amount of RTS/CTS packets decreases with a higher packet size, since less packets are needed to transmit the same amount of payload. Noticeable in both measurements is an unusually high peak in the region of “500 Bytes packet size with 5 Mbits/s CBR”, due to a disadvantageous combination of the two parameters [28]. This particular case requires an optimized tradeoff between UDP packet length and transmit bitrate. VIII. C ONCLUSIONS This work focuses on 802.11 based wireless mesh networks, deployed to transport real-time streams. As shown
0.5 Mbits/s Fig. 9.
1 Mbits/s
5 Mbits/s
20 Mbit/s
RTS/CTS Rate vs. Packet Size with HC=2
by the so far discussed flaws of 802.11, the development of network solutions that provide abundant streaming capacities for future rich multimedia formats is critical for the ongoing adaption process of 802.11. Especially home networks based on standard single channel 802.11 links may not be able to provide sufficient bandwidth resources and high quality links to satisfy the constantly growing quality-of-service demands. Taking this work as a base for such efforts, we propose a multi-layer approach. This allows routing metrics to get a more adequate view on the actual link states. Moreover, results of packet inspection methods can be used to facilitate traffic distribution over physically separated routes in the MAC layer, to decrease the general packet congestion per channel. A novel MAC layer instance to control the channel assignment and channel bundling of connected interfaces forms part of the multi-layer strategy. To optimize the mixture of MIMC and bundling approaches is an important task for the future. Now we are on the boarder to second and third generation mesh networks that offer quality features beyond simple connectivity of all nodes. 802.11s is clear evidence that mesh structures are finally accepted in a broad range of wireless applications. To help the IEEE 802.11 standard to evolve with future requirements, its network structures have to become more dynamic and able to fully exploit the channel diversity offered by the PHY layer. Enhanced spectral efficiency and adaptive multi-channel utilization are the keywords for a significantly increased network capacity.
R EFERENCES [1] Nouri Omheni; Faouzi Zarai; Lotfi Kamoun, Efficient Mac Layer Handoff Management Scheme for Wireless Mesh Networks, in Proc. of 17th IEEE International Conference on Telecommunications 2010. [2] G. R. Hertz; S. Max; E. Wei; L. Berlemann; D. Denteneer; S. Mangold, Mesh Technology enabling Ubiquitous Wireless Networks, in Proc. of the 2nd Annual International Wireless Internet Conference (WICON 06), Boston, MA, USA: ACM, Invited Paper, pp. 11, August, 2006. [3] S. H. Jeon; J. M. Lee; J. K. Choi, Performance Evaluation of Enhanced 802.11 in Wireless Mesh Networks, in Proc. of IEEE ICACT 2008. [4] A. Raniwala; T. Chiueh, Architecture and Algorithms for an IEEE 802.11Based Multi-channel Wireless Mesh Network, in the Proc. of IEEE Infocom 2005. [5] www.freifunk.net [6] Wireless Networking in the Developing World, http://wndw.net [7] Building a Rural Wireless Mesh Network, http://wirelessafrica.meraka.org.za [8] IEEE 802.11 WG, IEEE Std. 802.11, 1999 ed., Part II: Wireless LAN MAC and PHY layer specs. 1999. [9] Maaroufi, S.; Ajib, W. ; Elbiaze, H., Performance Evaluation of New MAC Mechanisms for IEEE 802.11n, in the Proc. of First International Global Information Infrastructure Symposium, 2007. [10] Chenzhe Zhang; Karol Kowalik; Mark Davis, An Experimental Study of the Impact of Using Multi-Radio in WLAN Mesh Networks, in the Proc. of IEEE WICOM 2009. [11] R.Draves; J.Padhye; B.Zill, Routing in Multi-radio, Multi-hop Wireless Mesh Networks, in the Proc. of ACM Mobicom, Philadelphia, PA, 2004. [12] J.Robinson; K.Papagiannaki; C.Diot; X.Guo; L.Krishnamurthy., Experimenting With a Multi-Radio Mesh Networking Testbed, in the Proc. of WinMee, 2005. [13] Chen-Mou Cheng; Pai-Hsiang Hsiao; Kung, H.T.; Vlah, D., Adjacent Channel Interference in Dual-radio 802.11a Nodes and Its Impact on Multi-hop Networking, in the Proc. of IEEE Global Telecommunications Conference GLOBECOM ’06, 2006. [14] Nikos Passas; Dimitris Skyrianoglou; Panagiotis Mouziouras, Prioritized support of different traffic classes in IEEE 802.11e wireless LANs, ACM Computer Communications Volume 29, Issue 15, September 2006. [15] Luciano Bononi; Marco Di Felice; Antonella Molinaro; Sara Pizzi, A Cross-Layer Architecture for Efficient Multi-Hop Communication in Multi-Channel Multi-Radio Wireless Mesh Networks, in the Proc. of 6th Annual IEEE Communications Society Conference on Sensor, Mesh and Ad Hoc Communications and Networks Workshops SECON 2009. [16] Ganesh Venkatesan, Intel Corporation, Multimedia Streaming Over 802.11 Links, IEEE Wireless Communications - Industry Perspectives April 2010. [17] Guido R. Hiertz; Dee Denteneer; Lothar Stibor; Yunpeng Zang et. al., The IEEE 802.11 Universe, IEEE Communications Magazine, January 2010. [18] A. Tonnesen; A. Hafslund; Ø. Kure, The Unik - OLSR Plugin Library, OLSR Interop and Workshop, 2004. [19] IETF. RFC 3561. http://tools.ietf.org/html/rfc3561, June 2010. [20] Sana Ghannay; Sonia Gammar; Fethi Filali; Farouk Kamoun, MultiRadio Multi-Channel Routing Metrics in IEEE 802.11s-based Wireless Mesh Networks - And the Winner Is . . . , in the Proc. of First International Conference on Communications and Networking, ComNet, 2009. [21] Liang Xu; Koji Yamamoto; Susumu Yoshida, Performance Comparison between Channel-Bonding and Multi-Channel CSMA, in the Proc. of WCNC 2007. ¨ [22] C. K¨ohnen; C. Uberall; F. Adamsky; V. Rakocevic; M. Rajarajan; R. J¨ager, Enhancements to Statistical Protocol IDentification (SPID) for SelfOrganised QoS in LANs, IEEE ICCCN 2010. [23] Christian K¨obel; Walter Baluja Garc´ıa; Joachim Habermann; Nico Bayer; Dmitry Sivchenko, Dynamic Channel Bundling in 802.11a based Media-Transport Mesh Networks, in the Proc. of IEEE Consumer Communications and Networking Conference CCNC’11, 2011. [24] Liang Xu; Koji Yamamoto; Susumu Yoshida, Performance Comparison between Channel-Bonding and Multi-Channel CSMA, in the Proc. of IEEE WCNC 2007. [25] M. Di Felice; G. Zhu; L. Bononi, Future Channel Reservation Medium Access Control (FCR-MAC) Protocol for Multi-Radio Multi-Channel Wireless Mesh Networks, in the Proc. of IEEE PE-WASUN’08, 2008.
[26] S.-L. Wu; C.-Y Lin et al., A New Multi-Channel MAC Protocol with OnDemand Channel Assignment for Multi-Hop Mobile Ad Hoc Networks, in the Proc. of IEEE ISPAN ’00, Dallas, pp. 232-237, 2000. [27] Ali Hamidian; Claudio E. Palazzi; Tin Y. Chong, Deployment and Evaluation of a Wireless Mesh Network, in the Proc. of the Second International IEEE Conference on Advances in Mesh Networks, 2009. [28] Frank Y. Li; Arild Kristensen; Paal Engelstad, Passive and Active Hidden Terminal Detection in 802.11-based Ad Hoc Networks, in the Proc. of IEEE Infocom 2006.
