Rottscher Rottscher

Roger P. Karrer

Senior Research Scientist

Deutsche Telekom Laboratories
Ernst-Reuter-Platz 7
D-10587 Berlin
Germany

Tel: +49 30 8353 58459
Fax: +49 30 8353 58409

Email: roger [dott] karrer [att] telekom [dott] de

Project Magnets

magnets

Introduction

I am the initiator, architect and principle investigator of Magnets - the world's most attractive network. (Magnets stands for Magenta Networks - magenta is the logo color of Deutsche Telekom). Magnets is a wireless infrastructure project to
  • investigate the fundamental properties of wireless access network, including high-speed wireless backbones and mesh networks
  • provide an experimental platform to evaluate the performance of network protocols and distributed systems
  • study user traffic and application behavior in a real environment, e.g. interactions of VoIP, real-time streaming and bulk data traffic
  • develop, deploy and evaluate architectures or components for a clean slate Internet design
  • integrate and study multiple wireless access technologies: GPRS, UMTS, WiFi and WiMax
Magnets is developed and deployed by a group of people at the T-Labs in Berlin, and it is used by a set of International collaborators: Core team:
  • Roger Karrer: PI
  • Harald Schioeberg, PhD student (of Anja Feldmann): routing and virtualization
  • Thomas Huehn, PhD student: trust; protocol performance
  • Doris Reim, undergrad student: our OpenWRT-Guru, mesh deployment
  • Prof. Anja Feldmann: "my very private steering and advisory committee" ;-)
International collaborations:
  • Prof. A. Wolisz, TU Berlin, Head of the telecommunications networks group (TKN)
  • Berthold Rathke: TU Berlin, Telecommunications networks group (TKN): our expert on lower layers
  • Prof. Antonio Pescape, Universita di Napoli Federico II: "our measurement guru"
  • Alessio Botta, PhD student, Universita di Napoli Federico II: Intern for protocol performance evaluation
  • Prof. Patrick Thiran, EPFL: analysis of 802.11 protocols
  • Adel Aziz, PhD student, EPFL: analysis of multihop 802.11 protocols
  • Prof. Jean-Pierre Hubaux, EPFL: fairness in mesh networks
  • Naouel Ben Salem, PhD student, EPFL: fairness in mesh networks
Past members and supporters
  • Istvan Matyasovski: intern. Routers setup and backbone configuration; performance testing; mesh node evaluation and testing.
  • Vivento Technical Services GmbH: antenna mounting and backbone setup
  • T-Systems International GmbH: backbone plannning and AP configuration
In the remainder of this document, you find the following information

Project overview

The aim of Magnets is to exploit the potential and the limitations of WiFi mesh networks. Given that first generation mesh networks, such as the MIT roofnet or TfA in Houston, have shown the feasibility to build city wide mesh networks, and seeing the increasing deployment of city meshes for commercial reasons, we are looking at 2nd generation meshes. The topics that we focus on are, among others:
  • multi-channel meshes: how can we scale the throughput by adding multiple WiFi cards into a device. What is the potential in an indoor setting, what are the factors and their impact in an ourdoor urban setting?
  • how do we build meshes that get as close as possible to their optimal capacity in practise? How do we build reliable meshes in practise? Is TDMA a solution in an urban setting? Which protocols achieve what performance at which layer - how much do they have to interact (cross-layer)?
  • how can we build self-X (X=healing, organizing,...) systems?
  • how can we virtualize a mesh network (given that we have multiple WiFi cards)
Thus, our focus lies on the experimental work, building on previous models and theory and trying to feed back our results for future models. The project is structured into 4 phases.
  • Phase 1 (May 2005 - April 2006): deploy a high-speed WiFi backbone and assess the per-link and multihop behvior of urban backbones. What are the characteristics of links over several hundreds of meters with directional antennas?
  • Phase 2 (May 2006 - April 2007): deploy a high-capacity indoor mesh network with 20 nodes (routerboards). This indoor mesh will be used to assess protocol behavior in a "controlled" environment and to test novel protocols before they are released to our outdoor mesh.
  • Phase 3 (May 2007 - April 2008): deploy a high-capacity outdoor mesh network with 100 nodes (avila gateways) at the campus of the TU Berlin. With the outdoor mesh we will assess the behavior of meshes in an urban area and assess how close to the maximal capacity we can get. Moreover, the mesh will be available to students at the university so that we assess the mesh charactersitics and protocol behavior with real user traffic.

Magnets backbone

The picture below shows the layout of the Magnets backbone (a pdf-version of the document is available here).

magnets

The Magnets backbone features an outdoor setup of a 5-hop link between the T-Labs at the Ernst-Reuter-Platz and T-Systems at the Goslarer Ufer in the heart of Berlin. All nodes reside on top of high buildings and therefore have unlimited line of sight. The end-to-end performance of the connections between these two locations is 108 Mb/sec physical speed. This speed is achieved in two ways: some links are deployed as 2 individual links, some links use a proprietary protocol (TurboMode) that doubles the speed by enlarging the frequency spectrum. In addition, two alternative paths exist between T-Labs and HHI - one direct path and one path via the TC building.
The link characteristics along the backbone vary in frequency and distance. The link distance varies from 330 meters to 920 meters. All link frequencies are in the free spectrum, some in the 2.4 GHz and some in the 5 GHz band.
Using this backbone, we want to investigate the link and end-to-end capacity over a long period of time. The variety in distance, frequency and protocol allows us to assess the performance in a variety of settings. As a true outdoor network, we are interested in average and variations of the link speed subject to controllable and uncontrollable parameters.

Magnets indoor mesh

The Magnets indoor mesh consists of 20 mesh nodes (routerboards) that can host up to 6 WiFi cards. The deployment in our Lab provides ample opportunities to place and modify the link characteristics. For example, on each floor, the nodes can be placed in opposite corners to avoid interference among them. Moreover, across floors, nodes see each other if they are placed on adjacent floors, but visibility is almost zero if the nodes are two or more floors apart. Finally, the indoor mesh can be combined with the backbone to assess multihop communication over topologies that can be even larger than 10 hops.

Magnets outdoor mesh

magnets

The Magnets outdoor mesh is an outdoor infrastructure-based wireless mesh network deployed at the TU Berlin. 100 wireless access points will be deployed at the campus of the university to provide ubiquitous high-speed wireless access to the students at the TU Berlin. Of the 100 access points, only some will be connected to the Internet directly - thus, they will forward the traffic in a hop-by-hop manner towards an access point with a fixed Internet connection. Traffic from or to the Internet can also be routed via the Magnets backbone.
We will strategically plan the deployment of the access points to assess the basic properties of a wireless mesh network, such as capacity, delay, fairness or scalability properties. Our access points will be equipped with up to 8 wireless cards to provide scalability, capacity and to study frequency planning in a mesh network. The cards will thereby operate in the unlicensed spectrum, both in 2.4 and 5 GHz range.
The mesh network will allow to experiment with a large variety of protocols and applications to assess the fundamental properties of next-generation wireless access networks. First, we will measure the fundamental properties of wireless networks at various layers. For example we will study fairness and performance trade offs in multi-hop multi-antenna systems over a long period of time and derive solutions at the MAC layer or joint protocols for MAC and TCP. Similarly, we will investigate scheduling algorithms at a low layer to support Voice over IP over multi-hop networks and assess the quality of Internet telephony by offering students access to our network.
At the transport layer, we will assess the effects of multi-antenna systems and optimized routing onto end-to-end performance. The increasing trend to optimize multi-hop multi-antenna systems or even different wireless access technologies leads to a high degree of packet reordering at the transport layer. Thus, the current TCP mechanisms are no longer suited for these optimizing access networks. We will fundamentally study the effect of lower layer optimizations and derive novel mechanisms that achieve high end-to-end performance. We will particularly focus on Internet scenarios where end-to-end performance may or may not include wireless systems, i.e., we will derive and experimentally validate protocols that achieve a high performance in wired and wired-cum-wireless systems.
Finally, at the application layer, we will assess the impact on application performance. We are targeting all applications from legacy bulk data transfer over peer-to-peer traffic to VoIP and real-time streaming of video. In particular, we have a joint project with the campus radio at the TU Berlin. We will stream the radio over Magnets and assess performance opportunities and challenges.

Inter-connecting meshes in Berlin

Berlin is a city of meshes. Besides Magnets (the youngest mesh in Berlin), the Freifunk mesh in the eastern part of Berlin contains 500+ nodes, and the Berlin roofnet is another experimental platform with exciting features. So, it is natural to ask the question whether these meshes could be combined. Challenges arise because the meshes have grown organically and are run by different adminstrations and have different protocol structures, e.g. routing or addressing at layer 2 vs 3 vs 2.5. Therefore, in a final phase, we are planning to interconnect these meshes and address these questions.

History and current Status

  • May 2005: initial idea of planning and deploying a wireless mesh at the TU Berlin. Decision to build Magnets in separate phases.
  • Jun-Aug 2005: Planning of the backbone. Selecting buildings, evaluation of hardware, feasibility, budget estimation.
  • Sep-Oct 2005: Putting the pieces together (location, hardware) and decision. Approval of the project by the management.
  • Nov-Dec 2005: Hardware purchase
  • Jan-Feb 2006: Configuration of the WiFi components in the lab
  • Mar-May 2006: Installation of the antennas and access points, testing and configuration.
  • May 2006-ongoing: measurements on the Magnets backbone
  • May 2006: Budget approval for the 20 nodes indoor testbed and student workers (manpower)
  • May 2006-ongoing: setup of the indoor mesh with routerboards, development and configuration of OpenWRT
  • May 2007-ongoing: measurements on the indoor mesh
  • Jun 2007: Budget approval for the 100 nodes outdoor testbed, including a PhD student, a system administrator and student workers.
  • Jul 2007: Invitation for negotiations for the EU Project CARMEN, a project dealing with several aspects of carrier grade mesh networks. This project will provide funding for manpower. Magnets is a potential testbed to experimentally assess novel protocols.
  • Jul 2007: Invitation for negotiations for the EU Projects Trilogy and 4WARD, projects dealing with clean slate Internet aspects. Magnets may be used in these projects to experimentally test and evaluate new ideas.
  • Jul 2007: Invitation for negotiations for the German Project G-Lab, dealing with a national experimental facility. Magnets may be used in this project as an experimental wireless testbed.
magnets magnets magnets
magnets magnets magnets
magnets magnets magnets

Pictures of the Magnets backbone. The first 3 pictures show the views: TLabs-TC, HHI-TLabs, TSI-ETF-HHI. The next 4 pictures show the antenna installations at TLabs, HHI, TC and ETF. TLabs has 5 antennas and 3 outdoor APs, HHI 3 antennas and outdoor APs, TC has 4 antennas and 2 outdoor APs, and ETF has 2 antennas at the wall. Finally, the last 2 pictures show the indoor mounting of the APs in HHI and TC.

Publications

Frequently asked questions

  • Where does the project name "Magnets" come from? Magnets is composed of two words: Magenta Networks. Magenta is the color of the Deutsche Telekom.
  • How does Magnets differ from other mesh networks Magnets shares in fact some similarities with other mesh networks. However, there are a few points that make Magnets unique:
    • Heterogeneity (technology): all Magnets mesh nodes have at least one 802.11g card. In addition, some nodes have up to 5 additional WiFi cards (802.11g and 802.11a), UMTS, Zigbee and Bluetooth for low-range communication and finally GPS. In the future, we may even add WiMAX cards once they become available.
    • Heterogeneity (capacity): The capacity of a mesh node changes as a function of the WiFi cards we put in (between 1 and 6)
    • Deployment: Magnets features both planned and unplanned deployment. The backbone and the campus mesh are carefully planned to reach high speed and high capacity. To extend the coverage of Magnets, we additionally give out nodes, thus leading to an unplanned deployment.
  • Were you aware of all the problems and issues of building such a network? Well, hell, no! Sometimes I am asking myself which devil rode me to plan and push this project. Thank goodness I didn't know all the obstacles - I probably wouldn't have started the project. Now, looking back, I am pretty proud of what I have achieved. For me, Magnets is really the world's most attractive network (but that's just a personal feeling).

Related projects

More to follow!