VoIP Performance in "MeshBed" - a Wireless Mesh Networks Testbed

VoIP performance in ”MeshBed” - a Wireless Mesh Networks Testbed N. Bayer, D. Hock, A. Roos, M. Siebert, B. Xu

V. Rakocevic

J. Habermann

Deutsche Telekom/T-Systems, Darmstadt, Germany {Nico.Bayer, David.Hock, Andreas.Roos, M.Siebert, Bangnan.Xu}@t-systems.com

School of Engineering and Mathematical Sciences City University, London, UK

University of Applied Sciences Giessen-Friedberg Friedberg, Germany

Abstract— This paper presents a WLAN based next generation wireless mesh testbed, called ”MeshBed”. The aim of this testbed is to investigate carrier-grade mesh network aspects from network operators point of view. The paper describes in detail the purpose, architecture as well as the features of the MeshBed. In particular, the Operations Support System consisting of monitoring and management tools as well as the integration into next generation network architectures will be discussed. Also the influence of wireless multi-hop communication on delay sensitive applications such as voice over IP (VoIP) is investigated by extensive measurements. It was found that wireless mesh networks are in principle capable of providing good quality support to VoIP applications. However, in highly loaded networks additional mechanisms are necessary to guarantee minimum requirements with respect to quality.

I. I NTRODUCTION Wireless mesh networks (WMNs) are increasingly becoming a viable solution for network operators to realise broadband wireless Internet access in a flexible and cost efficient manner. However, from network operators point of view such networks also need to fulfil carrier-grade requirements in order to deliver ”Triple Play” services with quality comparable to traditional access technologies. Recently a lot of research has been done in WMNs. Most of these activities are based on simulations. Simulations are important to investigate and optimise WMNs. However, to understand the fundamental parameters of WMNs, ranging from capacity constraints, scalable deployment and efficient management up to revenue potential, experiences and measurements derived from testbeds in realistic scenarios are essential. Currently a couple of those testbeds have been developed, e.g. Roofnet [1] or BerlinRoofNet [2]. However, only a few are known to investigate carrier-grade aspects, e.g. [3]. The contribution of this paper is two-fold. First, it presents the MeshBed which is a next generation WLAN (Wireless Local Area Network) based WMN, developed and deployed at T-Systems in Darmstadt, Germany. The paper discusses the purpose, architecture and the features of the MeshBed. In particular the focus is on carrier-grade issues like monitoring and management of large scale WMNs, AAA (Authentication Authorization Accounting) & security mechanisms, ”Triple Play” service provisioning and the integration into next generation 1 This work is partly supported by the German Ministry of Education and Research (BMBF)

978-1-4244-1645-5/08/$25.00 ©2008 IEEE

core networks. Second, VoIP performance is investigated based on extensive measurements within the MeshBed. Therefore, the most important VoIP parameters in terms of delay, jitter and packet loss have been investigated for a network with variable amount of background traffic. Results show, that in slight and medium loaded networks WMNs are able to support VoIP with good quality. However, in a highly loaded network, the quality of VoIP degrades significantly. The paper discusses the different sources of this quality degradation. The rest of the paper is organised as follows. Section II provides an overview of the MeshBed. Section III handles the integration of the MeshBed into a next generation network architecture. In detail, the integration into an overarching Quality of Service (QoS) and AAA environment is presented. Section IV presents results and performance of VoIP measurements and discusses current problems of WLAN in multi-hop as well as their influence on VoIP performance. Finally, Section V concludes the paper. II. D ESIGN ISSUES This section presents the MeshBed which is a next generation WLAN based WMN, developed and deployed at TSystems in Darmstadt, Germany. The purpose of the MeshBed is to investigate carrier-grade aspects from network operators point of view. Those aspects are: • Operations Support System for large scale WMNs • AAA & security mechanisms • ”Triple Play” service provisioning and the integration into next generation core networks • Performance analysis in terms of throughput, delay, etc. • Network based mobility support The MeshBed has been designed to offer services as known from other Broadband Wireless Access (BWA) networks: • Connectivity: the network should provide connectivity to legacy devices • Mobility: Users should be able to freely roam within the network but also to/from other networks • Integration: The MeshBed should be integrated with external networks. Especially with other wireless networks, e.g. WiMAX (Worldwide Interoperability for Microwave Access) or HSDPA (High Speed Downlink Packet Access) but also with wired networks, e.g. Ethernet or DSL (Digital Subscriber Line) to get access to the Internet.

2218

Currently, the MeshBed consists of 12 mesh nodes that are all deployed indoors, see Figure 1. For investigations in more realistic scenarios, it is planned to extend the MeshBed with a 15 nodes outdoor WMN. As hardware platform an embedded AMD Geode SC1100 Systems with 266MHz CPUs and 64 MB of RAM is used. For nodes that require more processing power, e.g. gateways, barebone desktop PCs with 3 GHz Intel Pentium 4 processors and 1 GB of RAM are used. All mesh nodes are equipped with Atheros Wireless Mini PCI WLAN cards as well as Ethernet ports and use operating systems based on Linux together with ”madwifi” [4], an open-source WLAN driver.

via the Ethernet interface of each MP for fixed UEs (FUEs) or wireless via the second wireless interface of the MPs for mobile UEs (M-UEs). This second WLAN card is configured in the access point mode for the operation in the 2.4 GHz frequency band (802.11g - already implemented in many mobile devices) and acts such as known from normal HotSpots. MPs with an access enabled WLAN interface are called Mesh Access Points (MAPs). Due to the separation of backbone and access network, the user terminals do not need any mesh specific functionality and off-the-shelf user equipment can be used.

Fig. 2.

Fig. 1.

Current MeshBed indoor deployment

A. Architecture The MeshBed architecture which reflects the principle architecture of WMNs is depicted in Figure 2 and consists of mesh backbone part and an access network. The mesh backbone is the core of the MeshBed since it provides routing as well as mobility functions and integrates the whole system with the Internet. The backbone is based on the multi-hop paradigm and is formed by Mesh Points (MPs), which are nodes that fully support mesh relaying, meaning that they are capable of forming an association with their neighbours and forwarding traffic on behalf of other MPs. High-speed wireless backbone links in the 5 GHz frequency band (802.11a) are used for communication between the MPs. Usage of the Atheros turbo mode with a max. gross data-rate of 108 Mbits is optional. Currently only one single channel is used for backbone communication. The 5 GHz band was selected to avoid the interference problematic of the 2.4 GHz frequency band in order to assure stability in the backbone. Mesh Point Portals (MPPs) in the backbone act as gateways between the MeshBed and other networks. Internet connection is provided via the Ethernet interfaces of each MPP. Backbone routing is done using the Optimized Link State Routing protocol (OLSR) [5] based on the implementation of Andreas Tonnesen [6]. The access network is used to connect user equipment (UEs) and thus users to the MeshBed in order to provide Internet access. The user connections can be either wired

MeshBed architecture

B. Features From network operators point of view, an Operations Support System (OSS) is essential in order to manage large scale WMNs efficiently and to guarantee a fast and easy deployment. A monitoring as well as management environment has been developed for demonstration purposes but also to gather experience in how to manage large scale WMNs. The OSS implementation can be seen as a framework which can be used to investigate various algorithms, e.g. for efficient distribution and collection of information. Besides the OSS also security is very important and appropriate mechanisms are realised in the MeshBed in order to protect user data from misuse and to increase trust and acceptance of the system. A brief description of these features is provided below. 1) Monitoring environment: The monitoring feature in particular plays a critical role to ensure 24/7 smooth operation of the network, which is an important characteristic of a carriergrade network. The system is based on a central server that collects all relevant data from the mesh nodes, processes these data and makes it available via a http (Hypertext Transfer Protocol) interface. To visualise the data a Java Applet frontend on the client side is used. On the server side a PHP backend together with a plugin for the OLSR implementation are used to gather the information. Furthermore, a SQL (Structured Query Language) database is used, to manage additional information of the nodes, e.g. node positions, node names, etc. In particular, the tool provides the following information: • Deployment of the network and position of nodes • Detailed information about every node including configuration, routing, status, CPU (Central Processing Unit) load, memory usage, etc.

2219

Detailed link information, e.g. quality, delay, etc. Routing information showing the route of every node towards the Internet 2) Management environment: The management environment helps to administrate large number of mesh nodes from a central point. In particular, the tool comprises the following features: • Distribution of configuration information from a centralised server to the mesh nodes • Group based as well as individual node configuration • Transfer of configuration parameters via https (http with ssl) to protect the information • Graphical User Interface (GUI) to ease the handling of the management software 3) Network access control and security environment: Security within the MeshBed can be subdivided into the protection of backbone communication as well as authentication and authorization in the access part. To protect MP intercommunication, two different approaches are implemented within the MeshBed backbone, which is assumed to consist of trusted MPs supplied by the operator. First, wireless link-by-link encryption for instance based on Wi-Fi Protected Access (WPA) or WPA2. Second, end-to-end encryption between MPs using tunnel connections, such as Virtual Private Network (VPN). Both approaches have their pros and cons and their performance in terms of additional delay, CPU load, overhead, etc. will be investigated within the MeshBed. The mesh access provides UE connectivity and is consequently responsible for user authentication and authorisation. From network operators point of view security in mesh access as well the aspect of user-friendly usage are important. However, user-friendly mechanisms often lead to compromises concerning security. Therefore, flexible mechanisms regarding application scenarios and user-friendly usage will be investigated. The following solutions have already been implemented: • Web-based network access control, such as known from HotSpots in cafes, restaurants or airports • Tunnel-based network access control, such as Point-toPoint Tunnelling Protocol (PPTP) • Wireless link network access control, such as WPA and WPA2 • •

III. M ESH B ED INTEGRATION INTO NEXT GENERATION NETWORK ARCHITECTURES

To make WMNs an integral part of operators access portfolio, an integration with the core architecture is required. Development and verification of possible integration concepts is also one purpose of the MeshBed. Concepts have been developed and implemented to integrate the MeshBed with overarching QoS and AAA architectures of the core network. Therefore, the MeshBed has been combined with the NGN (Next Generation Network)/IMS (IP Multimedia Subsystem) testbed that has been developed within the ScaleNet project [7]. This NGN/IMS testbed reflects a NGN network architecture which realises Fixed Mobile Convergence (FMC) and

allows users access to ”Triple Play” services independent of the access network that they are using. This is achieved by using the IP Multimedia Subsystem (IMS) within the control layer. An overview of the integration is given in Figure 3 while details will be discussed in the next sections.

Fig. 3.

MeshBed integration into NGN network architecture

A. Integration with overarching QoS mechanism Admission control (AC) is traditionally an important tool for network operators to provide QoS and to guarantee a minimum service quality in highly loaded networks. AC in WMNs is not as straightforward as in single-hop networks. Due to the multi-hop nature of WMNs, the service quality not just depends on one link but on all links along a route. For instance, the weakest link determines the bandwidth of the complete route. Thus, AC in WMNs should not only consider the load of the network but also conditions of all links forming the route. For instance, even if the load in a WMN is low, it might happen, that the requested service cannot be provided because the weakest link cannot support the required bandwidth. Compared to single-hop networks, additional intelligence is required to determine the capabilities of a route. An AC architecture has been developed. Figure 4(a) shows the general functionality of the developed mechanisms. The heart of the architecture is the Mesh Control Function (MCF). The MCF is implemented in all MPPs and receives QoS requests indicating for instance the bandwidth or delay requirements of a requested service. Therefore, the MCF interacts with the Resource and Admission Control Subsystem (RACS) which is an amendment of the IMS and acts as an overarching QoS control entity. Upon receiving a service request, created by a user, the IMS creates a QoS request towards the MCF. For debugging purposes these QoS requests can also be manually created and sent to the MCF via console. The MCF is then responsible for the evaluation of the requested parameters and to decide whether the desired service can be supported

2220

by the WMN or not. Reactive, proactive as well as hybrid QoS determination approaches have been implemented. The reactive approach determines the quality of a route only in case a request is received. This can be done with real-time measurements (e.g. using iperf, ping, etc.) or by estimating the performance using probing mechanisms (e.g. back-to-back packets). The proactive approach determines QoS capabilities all the time by leveraging information from lower layers. New routing metrics have been implemented in the routing protocol that allow to periodically broadcast these information throughout the WMN. Finally, the hybrid solution as a mixture of reactive and proactive mechanisms combines the advantages of each approach. This AC framework is used to evaluate different QoS evaluation algorithms in a realistic environment. Detailed information about the QoS integration approach can be found in [8].

the mesh central AAA (C-AAA) database. The hierarchical AAA approach comprises an additional hierarchical AAA (HAAA) database to enable temporal storage of AAA information within the WMN. This reduces network load. With other words, the WMN comprises at least two ADFs and at least two AAA databases, the central ADF (C-ADF) and CAAA as well as the hierarchical ADF (H-ADF) and H-AAA. All remaining MPs within the WMN comprise the ACF to terminate UE initiated network access requests and to provide network access control. Of course, all three approaches have pros and cons. The aim of the testbed is the investigation of these approaches and to analyse aspects like network load, delay, scalability, etc. in various scenarios, e.g. fixed networks without mobility, mobile network with many handovers, etc.

Mesh Access

Access Decision Function (ADF)

B. Integration with overarching AAA mechanisms Network access control, such as user authentication and authorisation as well as accounting, is a basic requirement for operator access networks. With focus on WMN integration into existing provider network architectures and fixed mobile converged networks an overarching AAA architecture is necessary as well. Thus, the MeshBed architecture takes into account the AAA framework Network Attachment Subsystem (NASS) [9] of the standardisation body European Telecommunications Standards Institute (ETSI) Telecoms & Internet converged Services & Protocols for Advanced Networks (TISPAN). ETSI TISPAN envisions NASS as network access control framework of their NGN architecture. NASS provides mechanisms for user authentication and authorisation, as well as device configuration (e.g. IP address). NASS delivers user policies to RACS regarding QoS configuration and interacts with IMS. The main components of the AAA architecture are the AAA database, the Access Decision Function (ADF) and the Access Control Function (ACF), as shown in Figure 4(b). User profiles with their policies are stored within the AAA database. The Termination Point (TP) terminates all access requests initiated by a UE. Depending on the used authentication mechanism, such as web-based login, wireless link encryption or tunnel connection, the ACF translates the access request into ADF compliant format using protocols, such as Remote Authentication Dial-In User Service (RADIUS) [10] or Diameter [11]. Furthermore, the ACF is responsible for gathering accounting data. The ADF carries out user authentication and authorisation considering user profiles within the AAA database and informs ACF about user authorisation status. Depending on this authorisation status ACF controls network access. In case of successful user authentication and authorisation ACF grants network access. The mesh AAA architecture can be realised in a centralised, hierarchical and distributed manner. Currently, only the hierarchical approach is implemented as it provides most flexibility concerning performance and configuration investigations, see Figure 3. Basically all AAA information is stored within

AAA

Access Control Function (ACF) Termination Point (TP)

(a) AC Fig. 4.

(b) AAA AAA and AC implementation

IV. VO IP PERFORMANCE MEASUREMENTS A. Reference Measurements in an Undisturbed MeshBed The VoIP performance measurements aim to cover a maximal range of WMN aspects to detect and to analyse possible problem sources. To make this possible extensive reference measurements in an undisturbed MeshBed not only with VoIP traffic and turbo mode enabled are presented first: 1) Constraints of the nodes’ WLAN interfaces and hardware: Maximal bandwidth and minimal delay of the MeshBed components have been analysed by one-hop connections working under ideal (radio) conditions. The maximal possible bandwidth has been determined to be about 30 Mbps. The minimal delay reachable between two neighbour nodes was measured as about 0.28 ms. 2) Minimal load of signalling and routing messages: Depending on the number of neighbours a MP receives on average between 400 byte, about 3 to 4 packets, and 2000 byte, 15 to 20 packets, of OLSRd signalling messages per second. Even the highest measured signalling bandwidth of 2 Kbps can be ignored in any network load. The signalling load issue is thus not assumed to be a problem in the MeshBed. 3) Dependency of different services on the route length: Changes in delay, loss and bandwidth due to an increasing number of hops have been quantified. Table I shows the results of measurements for one- to five-hop routes. The packet loss in

2221

30

avrg delay 0.2885 ms 0.7485 ms 0.9225 ms 1.3737 ms 1.6775 ms

min delay 0.2195 ms 0.3625 ms 0.458 ms 0.7825 ms 0.8895 ms

max BW UDP 29.0 Mbps 24.9 Mbps 18.0 Mbps 9.99 Mbps 7.53 Mbps

20 10

stdIPD (ms)

0

dist. bandwith (Mbps)

#Hops 1 2 3 4 5

loss (%)

the undisturbed MeshBed has been determined to be 0 % and is thus not listed in the table. The depicted values have been obtained from different source-destination combinations for every hop count and several repetitions of the measurements. This way influences other than the increasing hop count as for instance network interface speed or physical location of the MP could be minimised. max BW TCP 25.5 Mbps 21.2 Mbps 13.5 Mbps 6.97 Mbps 5.73 Mbps

0

200

400

600

800

1000

1200

0

200

400

600

800

1000

1200

0

200

400

600 time (s)

800

1000

1200

6 4 2 0 25 20 15 10 5 0

TABLE I D EPENDENCY ON NUMBER OF HOPS

Fig. 5.

The outcome of the reference measurements has shown that VoIP in undisturbed WMNs is possible at acceptable quality. Even for a hop count of five hops, the maximal measured values for the resulting end-to-end delay did not exceed a threshold of t = 5ms while at the same time the corresponding jitter was below 1 ms. B. Sensitivity of VoIP Traffic to the Network Situation The second part of the VoIP performance measurements investigates the sensitivity of VoIP traffic to an increasing load in the WMN with turbo mode disabled. Due to mesh inherent properties different sources of quality degradation are possible: These include among others packet collision on the air interface as well as overloaded queues in any of the MeshBed nodes. Both these influences could be clearly made visible by measurements in reference scenarios. Figure 5 shows the results of a measurement in one of these scenarios, the scenario of Figure 2 presented in the paper. A VoIP traffic from node A to D over three hops A-B-C-D is disturbed by traffic on a one hop route E-F in range of B-C with linearly increased bandwidth. In this case the problems are supposed to occur by packet collision on the air interface. The abscissa shows the time of the measurement in seconds. The ordinates of the two upper graphs contain the values of packet loss and standard deviation of the inter packet delay at node D over the measurement time. The lowermost graph’s ordinate shows the increase of the disturbing bandwidth from E to F during the measurement series. The curves mirror a practical quantification of the theoretically expected problems on the air interface. Obviously in this scenario the packet loss of the VoIP connection is influenced more by this disturbing traffic than the jitter. An increasing number of packets is thrown away, when the sending attempts fail. The packets that are delivered though still arrive without any bigger change of inter packet delay. Another perception is, when observing the first 500 seconds of the measurement series, that up to a certain level cross over traffic in a WMN has no significant influence to VoIP connections.

Influences of Cross Over Disturbers

V. C ONCLUSIONS This paper presents a next generation WLAN based wireless mesh network called ”MeshBed”. The objective of the MeshBed is to investigate carrier-grade aspects of WMNs. Therefore, several features are implemented to understand their fundamental behaviour and characteristics in realistic environments. Being part of the Triple Play service bundle of next generation networks, VoIP was initially chosen as the reference service within the MeshBed for extensive measurements. The findings are, that in general VoIP can be supported with good quality. However, in highly loaded networks quality drops due to different reasons and additional mechanisms are needed to overcome these problems. Objectives for future work are the implementation of QoS and mobility concepts. R EFERENCES [1] “http://pdos.csail.mit.edu/roofnet/doku.php.” [2] “http://sar.informatik.hu-berlin.de/research/projects/2005berlinroofnet/berlin roof net.htm.” [3] R. Karrer, P. Zerfos, and N. Piratla, “Magnets - a next generation access network,” in Poster at IEEE INFOCOM 2006, April 2006. [4] “http://madwifi.org/.” [5] T. Clausen, P. J. (editors), C. Adjih, A. Laouiti, P. Minet, P. Muhlethaler, A. Qayyum, and L.Viennot, “Optimized Link State Routing Protocol (OLSR),” RFC 3626, October 2003, Network Working Group. [6] “http://www.olsr.org/.” [7] M. Siebert, B. Xu, M. Grigat, E. Weis, N. Bayer, D. Sivchenko, and et al., “ScaleNet - Converged networks of the future,” Information Technology on IP based mobile Networks, September 2006. [8] N. Bayer, A. Roos, R. Karrer, B. Xu, and C. Esteve, “Towards Carrier Grade Wireless Mesh Networks for Broadband Access,” in Proc. First IEEE International Workshop On Operator-Assisted (Wireless Mesh) Community Networks 2006 (OPCOMM ’06), Berlin, Germany, September 2006. [9] Telecommunications and Internet converged Services and Protocols for Advanced Networking (TISPAN). Network Attachment Subsystem (NASS). 2007. [10] C. Rigney, S. Willens, A. Rubens, and W. Simpson, “Remote Authentication Dial In User Service (RADIUS),” IETF Standard RFC 2865, June 2000, network Working Group. [11] P. Calhoun, J. Loughney, E. Guttman, G. Zorn, and J. Arkko, “Diameter base protocol,” IETF Standard RFC 3588, September 2003, network Working Group.

2222