What we can build and what is useful for the community? ORBIT or Emulab is one instance, but still needs for multiple services parallelly, e.g., SDR, only one instance running. Well we can have multiple DSPs running parallelly. But there are complexity and cost issues. By bootstrapping, using different frequencies, separate control path.
Need practical recommendations.
Specific discussions to each question:
Q1: What are the emerging technologies with high impact on network design?
Needs: Wideband, higher performance ADC/DAC. What‘s the radio technologies networking people care about?
(1) Cognitive Radio (2) MIMO (3) WIMAX (4) 802.11.x (5) Sensor radios (6) 3G, 4G (7) UWB.
Why put 802.11? It’s cheap and easy to deploy. (1) is related to flexibility and spectrum, (2) (3) are longer range and capacity. (4) is about speed at short distance, cost. (5) is about low power lower bit-rate (6) is optimized for wide area. (7) is about short range higher speed.
Need recommendations for a deployment in 2 years. Concentrate on (6) since currently we do not have much testbed on it. Also sensor network. Currently no open radio API yet. Modularity of radio components.
Which has the most the impact on internet? 802.11, low cost. Categorize them to room-level, city-level, etcl. Internet research, needs micro-level radio, and macro-level radio. (1) and (2) are more emerging.
Open cellular radio. How to open the cellular interfaces? Build something for community.
Modular: open-interface. Have one of the companies to open it. High density, easy to use, PnP. How to improve it? Deisgn on chips? SDR? Open 802.11 firmware; some 3G/4G radio, 802.16; sensors (mote, zigbee), open 802.15.4; Future: CR, agile radio.
Q2: What wireless net platform is available, which one is suitable for a potential new experiment infrastructure?
List of available platforms:
- Clients; radio AP’s, BT’s (modular, open, virtualizable);
- Radio routers/forwarding nodes
- Measurement equipment
- Sensor gateway
- SDR, 2 yrs to releases to use.
Mote is useful but limited? What could you do in research with this devices? Need a lot new applications, e.g., location service, information delivery on the move, etc. Large area coverage is so controlled by companies now and there is new need to bring new stuff. See radio infrastructure happens. Using our own stuff, either mesh-net or wide-net.
Continue on platforms, e.g., Intel stargate:
(1) sensor radio (2) AP can be a PC (3) wireless router (support multiple radios) (4) is Cognitive radio. (5) sensor gateway (Not easy to do fair queuing in pc).
Q3: How is programmability and flexibility best included in a new experimental network?
Tools: matlab, simulink, tools chain, Geni radio,
Design for cross platform use. User space programming/kernel space programming, embedded progress programming, etc.
Do we have space-shared sensor network. We have management software. Multiplex this stuff, resource allocator. Running multiple clients on testbed in buildings, you can share. Need to coordinate with the frequency. How to quantify and allocate resource?
- tools/design flow
- linux platforms
- spatial shares (for sensors)
- spectrum coordination: measure, quantify, allocate
- control/monitoring framework
- Virtualization possible at APs BTS, not at sensors.
Measurement at outdoors is expensive. Spectrum analyzer, do real time, Spectrum sensors (<$10K) is cheaper but with limited functions. Expensive on wide-area coverage analyzer.
Topics about virtualization: need a control layer. Do migrating.
Spatial vs. temporal virtualization. Using the same MAC. Research on L3 above. Build wireless virtualized network. Building zero-MAC on testbed.
- PlanetLAB: virtualization model: slices, control and management.
- Harmonizing control access wired and wireless.
- Coordination between wireless testbed, planetlab, sensornet,…
Q4: Concepts for a useful large-scale experimental network for sensors? For cognitive
Testbed can help protocol design but need some outdoor deployment. Real-time traffic requirements.
Need to share testbed facility between users.
Components related to new Internet:
- merging radio tech – demonstrator network
- network wide-area mobile service
- sensor deployment
- radio grid + zigbee
- adhoc mesh for metro or sensor net
- radio infrastructure
- location service and other services inside Internet
John: Can testbed like Orbit help some protocol of sensor network design if it does not use real field data.
I think real traffic is important... testbed is different from field deployment for sensors from real biologists, etc
Larry: we can give them some sense of protocol design from the usage of testbeds.
John: I see two kinds of sensor testbed. One is for protocol design. The other is for scientists (which utilize sensor networks)
Virtualization allows us to share testbed facility between real users and ...
Jay: Sensors deployed in this testbed could attract multiple scientists....
John: To reach a broad set of sensor network researchers, it is better to have a real large network.
Ray: what is the tool we need to allow low power sensor access? Toolkit is good. Without that, people cannot start.
Krishan: future radio tech and existing radio needs to be separate
Ray: Something like ad-hoc mesh need new clusters... radio infrastructure is another cloud.