1. Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 1 A Geography-Aware Community Wireless Testbed Bow-Nan Cheng, Max Klein Shivkumar Kalyanaraman Email: shivkuma@ecse.rpi.edushivkuma@ecse.rpi.edu : “shiv rpi” RPI, Troy, NY Funding: NSF-ITR 0313095, Intel

2. Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 2 Outline q Wireless Ad-hoc Mesh Networks: Challenges q Physical Layer: A “street-level” network q Network Layer: Addressing Framework and Auto- configuration q Microcosm Test Bed Lab q Future Work

3. Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 3 Community Wireless: What’s New? q Lots of work in ad-hoc networks coming to fruition q Startup companies in the mesh networks space: q Tropic networks, Mesh networks etc q What’s different in CWNs? Why a testbed? q Unmanaged, but operational network q Ad-hoc, but operational q Fixed (I.e. not mobile) q Medium scale q Supports organic growth and evolution q Supports legacy internet traffic

4. Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 4 Community Wireless Networks: Challenges ? Why a Testbed? q Auto-configuration and auto-management: q Only lightweight “governance” allowed q Beyond addressing: routing, naming, other protocols q Higher quality than ad-hoc networks to support legacy applications (links, end-to-end transport) q Capacity maximization (c.f. Gupta/Kumar results) => routing choices, traffic engineering q Cheap, simple, standards-based components. q New community apps (eg: p2p video, games). q Several nitty-gritty issues leading to new protocol design challenges… best researched in a testbed…

5. Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 5 Broadband exists. Why CWN? q Ans: Multiplicity. q Cable modem and DSL and CWN and … q Commodity => cheap to get multiple access facilities … Ethernet WiFi (802.11b) 802.11a USB/802.11a/b Firewire/802.11a/b Phone modem

6. Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 6 Physical Layer: Street-level vs Rooftop q Leverage directionality and quasi-LOS of streets => better quality links q Omni-directional vs. Directional Antennas Eg:

7. Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 7 Antennas: Pringles cans disappoint q Cheap … q But low gain: 6.01425 dBi

8. Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 8 COTS Directional Antennas 2.4 GHz 12 dBi Radome Enclosed Yagi Superior performance Light weight All weather operation 45° beam-width Can be installed for either vertical or horizontal Polarization. Includes tilt and swivel mast mount

9. Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 9 Network Layer: Addressing q Goals: q IP-based architecture q Support for geographic routing q Why? Medium-scale, Organic Growth, Location- services q Support for local traffic engineering q Why? Dirn antennas, capacity maximization. q Support for distributed auto-configuration q Proposed Addressing Framework: q Geographic Distributed Addressing (GDA)

10. Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 10 Geographic Distributed Addressing (GDA) Idea: Hash GPS -> IP address

11. Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 11

12. Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 12 Why GPS-to-IP? q GPS-to-IP => q Medium-scale, Server-less auto-configuration, q With routable addresses… q IP address has dual semantics: geographic and topological q Long-range geographic routing/TE q Short-range: (k-hop) topological RF-aware QoS routing q … using the same IP/geographic address… q Location information also leveraged to auto-configure other L2/L3 protocols: q Eg: cluster/area boundaries for routing, support location-based CWN services

13. Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 13 Future: Geographic Source Routing Source Routing i.e. flex, large header Greedy Routing i.e. no header or state, but no flex Trajectory-Based Routing Gives us ability to pick randomly from a large number of physical routes

14. Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 14 Deployment Plan q Microcosm lab: internal prototyping, tests q Logistics for long-term deployment around RPI campus (eg: dealing with landlords etc) q RPI-CIO providing access points at borders of campus to help connect the CWN to RPI’s internal network

15. Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 15 Microcosm Test Lab Variable attenuators, and directional antennas allow flexibility in testing We also intend to use public facilities like Utah’s Emulab Wireless

16. Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 16 Testbed Hardware… contd

17. Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 17 Testbed: Software Fixed Nodes: - RedHat Linux Kernel 2.4.20-30.9 - Click Modular Router: CVS March 5 Autoconfigured Nodes: - Redhat Linux 9: 2.4.20/21 Kernel - Click Modular Router: CVS March 5 - HostAP 0.2 Driver + Utilities for Firmware flashing - GPSD 1.10 - DHCP Server + preconfigured NAT w/ iptables ipmasq - Customized GSP Autoconf Scripts - Webserver + SSH in the future -Intel Stargate/east platform will be considered as well

18. Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 18 Future Work q GeoNet Framework to verify intersection between Geography and Topological routing q K-hop RF Awareness, QoS routing q Distributed Geographic Traffic Engineering q Test Bed Improvements q GPS Simulation q Lab Reflectivity q Transport and application-layer activities kicked off q Key: survival & quality under heavy erasure conditions. q Collaboration ongoing with Intel and AT&T Research

19. Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 19 Vision #1: Multipath P2P Video/Data Over CWNs “Fast” path I “Slow” path P Traffic engineering & Transport level upgrades

20. Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 20 Vision #2: free-space-optical CWN Ongoing NSF-STI project…

21. Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 21 Thanks! : “shiv rpi” Student Heroes: Bow-Nan Cheng: chengb@rpi.educhengb@rpi.edu Max Klein: maxklein@stanford.edumaxklein@stanford.edu