Future Internet Architecture & Protocols and Integration of Wireless/Mobile/Sensor Networks
Waveform agility and the Next Generation Internet
Three disruptive trends
1. Increasing waveform diversity – MIMO, UWB, OFDM etc
Each creates a different types of link profile.
2. Software ->Agile -> Congnitive radios; to adapt – switch between radios
3. Ad hoc /Mesh Networks: 4-5 standards activities many vendors; trend towards increasing presence.
Confluence of three disruptive trends
Links will become part of the topology; Then a link is no longer just an input but also a parameter;
The impact of this:
Advantage – “Transit networks” may use multihomed mesh network to replace a thin wired-Internet pipe when needed. The aim could be that of eventual connectivity where packets are delivered when possible – when topologies unite to form a connected network.
Instead of abstraction and hiding details, must exploit physical layer flexibility – cross layer and beyond to new layering. Transport protocols should incorporate feedback. A policy rulebook is needed to control behavior based on machine encoded policies.
Experimental testbeds need to consider reconfigurable hardware.
Question: Victor Bahl: build adaptation platform over radios; or build configurable radios; This was identified as an important point to discuss during breakout session.
David B Johnson
Mobile IP and adhoc networking in the Internet
What’s right and what’s wrong about mobile IP ? Right: Very scalable all state at end points (one mid point) and other points. What’s wrong: Registering is too slow; both v4 and v4 hurt by dependencies.
Right - lot of work on unicast routing; Wrong ? Need lot more work on multicast and other kinds of routing, standard services like DNS need to be provided; many other issues.
Missing pieces to consider:
1. Interconnection of new emerging adhoc networking technologies, internet and mobile IP. 2. Support for mobile routers. 3. Session or connection persistence. 4. Addressing, mobility and auto-configuration – doing autoconf with adhoc is a hard problem.
Killer app for adhoc apps is still unknown.
Question: what percentage of people use mobile IP.
Very small. But cell phone industry is very interested in IP
Title: (Phy-Link) Layers in a brave New World
The “link” today is vanishing due to the various wireless technologies. The physical layer has maximum performance. The performance diminishes at higher layers due to overheads.
Broadcast to all neighbors means what ? all channels etc ? cross layer design needs precise definitions.
There is a vanishing boundary between regions
Cooperative diversity may be utilized at higher layers. Cross layer adaptations are required. Do we need to have a meaningful separation of layers ?
Title: Networking in the Small Lessons for the Internet–at-large
The network today is no longer “dumb”. There is significant network infrastructure between end hosts. There are a variety of access technologies. Inter-networking of access networks to seamlessly move a session is required. This requires a new network framework that supports cross layer information exchange. However, upcoming enabling technologies still assume a “smart host, dumb network” model.
We envision bandwidth auctioning, which is a bid for additional bandwidth based on demand. These trends need addressing in future Internet. Issues that need addressing include scaling and load balancing.
Wireless X-ities for wireless mobile/sensor networks
Identifying what exactly is the problem is a significant portion of the problem;
Need to route message informing of the problem to the right user.
Alleviate problem by
- introducing helper class in the protocol stack informing layers of protocol status;
- nodes communicate with one another of status
For Fault detection and correction, need to exploit model of an existing working system.
The talk covered other aspects of Network management. Please refer to the slides for complete details.
Rajive Bargodia raised question whether network management is a valid NSF research problem.
Title: Architectural Implications of Multi-Dimensional Scalability in Wireless Networks
There is a desire for multi-dimensional scalability to include diverse node technogies, applications, nature and physical reach of subnets, security and management models etc. There are several challenges to operate Wireless as an Inter-network. Its implications include subnet “interoperability”, dynamic spectrum sharing among others. For subnet interoperability, the complexity actually lies in the control plane. Better spectrum efficiency is achieved with Agile spectrum use. Better management and security in wireless systems requires them to be self managed, identify rogue entities better among other aspects.
For discussion need to include defining the scope of new architecture
Wade trappe: Typically, cross-layer optimization is considered from single-user point of view, but recent results indicate that it can cause problems in a multi-user scenario where everyone is trying to optimize their own issues (game theory studies have shown this, for example). Then, throw in security and you have ways for people to cause problems (greedy behavior). Peter agreed and said that this is an issue to be concerned with.
Vehicular networks (VANETs) are an important application of wireless networks. They require location awareness of the protocol stack. The application requirements for a location service need to be identified. However with no such application available today, it is difficult to study requirements. The re is no coherent network architecture to study location service. It’s a chicken and egg problem.
We need to bootstrap support for location-awareness in the new experimental testbed system. Privacy architecture components such as for access control, on-device localization and a concept of location cloaking could be used ; Consider what an overlay service to collect islands of ad hoc service look like.
Title: Requirements of next generation Wireless Internet
Wireless extensions of the Internet will extend far beyond the number of humans or computers, thanks to RFID tags and wireless enabled sensors. RFIDs may be used to monitor and care for the elderly, animal health monitoring (A2M) etc. these enable sensor nets; there is also vast data availability.
Large number of nodes are a typical characteristic of sensor networks. However, end-to-end capable nodes may be undesirable. Spatio-temporal diversity for functionality may be used for network manageability.
Notions of pricing and fairness in a resource-constrained environment must be reconsidered.
For network management, the approach could be of a loosely structed p2p learning framework; Each node uploads its experiences to a common space from which other nodes learn lessons. Need to learn from people’s experience instead of removing people from the loop. Napsterization of the control plane may be considered too.
Question: Dipankar Raychaudhuri was curious to understand the learning time frame granularity.
Answer: Learning could be on a coarse level as well as on a fine level. Info may be uploaded into into learning-cloud at the end of the day from which other nodes learn.