Emerging Technologies and Experimental Infrastructure Concepts
ORBIT Testbed Experience & Experimental Platforms for Future Networks (Max Ott)
The action is to minimize the cost of edge access and push up the network abstraction such as semantic routing. Why testbed? The real world does not fully trust simulations, real world isn't perfect and real world is too messy, hard to capture all externalities. Then introduce Orbit Grid and its architecture. Importance of reproducible results! Define, experiment, and collecting results. Some practical considerations: large scale testbed, investment, shared, require more engineering efforts. What's a good testbed node? Costs: keeping it running, keep it current. Some challenges: software rados, MAC layer (open source, etc.).
(1) Where to deploy the testbed? Create a testbed, how to distributely deploy it? There are different complexities for different environment deployment. Characterize different scenarios. There are different phases of an experiment.
(2) Is there any analysis tools? How to track the measurements?
(3) Cross-layer abstraction: how to control lower layer? Protocol performance depends on lower layer parameters.
What is wireless overlay? (Dirk Grunwald)
The goals of adopting overlay network is to provide basic transport (of some kind) to enable new services, allow novel protocols / networks and eliminate QoS concerns. What is wireless overlay? Using existing bit-pipes to enables new networks services. Most wireless network problems are performance and QoS related. Overlay models assume inflexible infrastructure. Wireless net need exploration, not retrenching. Example of PHY layer choice: no explicit control vs. power control vs. rate control, and adjusting broadcast rate or power to force local routes. SoftMAC: software radio on the cheap. MultiMAC - multiple simultaneous MACs: reuse component infrastructure to use multiple mac layers, moving to multi-radio software radio architecture. Much of the SDR impact can be below network layer: etiquettes for BW, negotiation allocations
(1) Challenge: considering 10% low utilization rate now, we can have spectral reuse, better modulation, refarm many technologies for higher spectral efficiency. And different architecture and ways to reuse the resources.
- wireless, not like wired, is too limited.
(2) What is the capacity of the last mile. How to merge wireless/wired overlay?
4.3 Barriers for Disruptive Network Innovation - Greater for Wireless or Wired?
Emulab/Netbed Experience (Jay Lepreau)
Wireless has limited bandwidth. Link layer is trivially enabled by SDR (USRP, WoftMAC), but they are mobile and untethered. Testbed offering real marginal values to users are more likely in wireless. Wired can only offer bandwidth or security. Regulatory and legal issues: limited availability of open spectrum; licensing barriers. But there is more open spectrum coming. Emulab: big: 400,000 LOC growing ~80K/yr. Emulab is a portable, device independent OS for controlling testbeds of network devices.
(1) How is it useful. Nobody use it if too complex.
(2) Most publications don't have experimental results.
SIP/SIMPLE as an overlay protocol for the Internet (Arup Acharya)
SIP: session initiation protocol. Impact of SIP in the enterprise: most applications are now by http. Impact in telecom: circuit-switch -> SIP-based IP communications. SIP in 3rd G mobile networks: IMS. SIP messages: headers + body. Capture the best of both: efficiency of IP headers, flexibility of application layer messages. Impact on internet: A new de-facto control layer based on SIP. Scaling SIP servers. SIP on wearable devices (linux watchpad).
(1) Is SIP good for everthing? Can we look at different ways to embed control in a SIP-like protocol, Is it efficient? Notion of the name. Definition of IP is misleading. Signaling protocol. SIP is based on messaging control. Use a common protocol.
Cognitive/Programmable Radios (Gary Minden)
Agile radio drivers: RF systems, ADC/DAC, digital processing, new radio architecture; Economic: user service demand, new market models, spectrum stewardship; National policy: national security, homeland security, public services. Radio resource space (ElectroSpace): 4-dimensional resource space: time, frequency, space, signal format/coding. Receiver oriented electrospace: the space defined in receiver; receive state distributed to network. Cognitive challenges: self-configure radios; Mission oriented radio configuration. General radio model: compression, encrypt, error control, spreading, modulation. Channel: thermal noise, mutual inter, jamming, multipath, .
(1) When will agile radio is mature for higher layer research? A: Next summer to release to community. Takes a year or 9 months.
(2) We want a platform to support a variety of things: change modulation scheme, spectrum management. Programmable radio on a higher layer protocols. Improvement on bits/Hz, etc..
Technologies for Converged Networks of the Future - SoftRouter and Base Station Router as building blocks for future systems (Krishan Sabnani)
Tomorrow's converged network: integrated network management, security, reliability; QoS enabled packet core network. Access: provide homogeneous bit stream to internet. Base Station router: push intelligence to the edge. BS router terminates all air interface; simplify network with better QoS. Core: softRouter: enable carrier grade performance. There are 3 key components: separate complex control plane processing functions from the transport plane; control plan processing functions on dedicated external control plane service; standard interface. Benefits: lower costs, new applications, better scalability, enhanced reliability, increased security.
The WHYNET Testbed (Rajive Bagrodia)
Potential impact: cross layer interaction. Approaches: hybrid testbed; heterogeneity in wireless tech; hybrid testbed. Combine realism of physical, emulation, simulation. A case study: adaptive video streaming performance in MANET; qualnet simulation framework; impact of real traffic model: CBR: AODV and DSR perform equally well, but for sensor network traffic model: DSR performs 500 times better than AODV. Testbed challenges: scalability; application centricity; interoperability; cost, robustness, efficiency.
(1) DSR 500 times better AODV: what is the problem and actual traffic model. Simulation models usually do not capture many real behaviors.
(2) Is scaling is a problem?
Sensor Network Infrastructure: Re-created to Rarefied (John Heidemann)
Ideas are always reusable. When is more than the idea reusable? Code, hardware, testbeds, traces, datasets, models. Rarefied Components: centralized and shared. Large: emulab, orbit; physically distributed testbeds: Dartnet, Cairn, planetlab. Where from here? SensorNet community needs some rarefied components.
(1) About millions of nodes. Why would we build a million sensor node network? Hierarchical? Why?
We have million node internet. Interesting to have million nodes simulations.
Large-scale Edge Services: 1 million lines of code, 1 million devices, 1 week (Anthony Joseph)
Novel classes of edge services: . (a list). Challenge: develop and deploy a 1MLoC application in one week on 1M nodes. It must be easy and simple to experiment with and to evaluate resulting distributed applications. Opportunities for novel edge devices. Cross-layer and cross-network services. Extensible xml vs. IP hourglass. What should be pushed into a new Internet? Command and control tools for apps and network. Reusable state management, comm., comp., & coordination tools. Security and privacy: identify useful primitives' information that infrastructure can provide and collect, e.g. packet level authentication. Testbed: O(1000) real nodes, wirelessly connectable to every other node.