June 2007CRI workshop (Boston, MA) Testbeds Henning Schulzrinne Columbia University
June 2007CRI workshop (Boston, MA) Questions to get started What are the fundamental networking questions? –content distribution - distributed resource –“new architecture”: airport model –> 4 billion devices: does IPv6 work? –deploying new services (multicast, QoS) –distributed data stream processing - computing in the core –application-sensitive (video); dyn. resource allocation; isolation –tolerance to (say) long latencies –security issues: network or not? –disconnected operation - network support –low-power networking
June 2007CRI workshop (Boston, MA) Testbed goals Testbed as an end, rather than (just) means to another end Experiment with virtualization Create smaller self-contained domains Restructure the industry
June 2007CRI workshop (Boston, MA) Why testbeds? Which of these can be answered by testbeds? –experimental system: build, evaluate, write PlanetLab, Emulab –deployment platform (e.g., Coral) to be used by people geographic scale (lots of users can reach it) –scale for routing? --> probably not –“show me” (industry, other researchers) –deploy things that are not yet mature
June 2007CRI workshop (Boston, MA) Why testbeds? Can’t build it at one institution –cost –scale (distances, nodes) –maintenance (technicians, programmers) Can’t justify at one institution –cost vs. usage intensity
June 2007CRI workshop (Boston, MA) Testbed models “Internet3” –focus mainly on L2, L3 –high-performance, low-latency testbeds Testbeds from commodity parts (e.g., routers) –subsystem components –transfer of technology to “real” routers? (cf. Linux model) –power as observable quantity Federation of institution test beds –“cooperative” model (mutual resource sharing) PlanetLab on steroids –resource-limited experiments (memory, CPU): who gets what? user count? needs resources -- tiny right now –wireless motes, ORBIT: sign up for time Kits and tools
June 2007CRI workshop (Boston, MA) Testbeds and users aware vs. oblivious users –old Gigabit testbeds, DARTnet... Internet2 roll out applications without access to actual testbed (Internet access) performance isolation
June 2007CRI workshop (Boston, MA) Testbed desirables Representative testbed –large potential user population –running non-Linux for networking Need programming model –language abstractions –data flow style, scheduling