Breakout session

Wireless/Mobile/Sensor Network Requirements (Coordinator: Mario Gerla)

Session Notes

We will identify the requirements using three scenarios:
1. The individual mobile user, interacting with Internet resources only
2. The mobile ¡°constellation¡±: the user equipped with several devices, interacting with the Internet, with environment and with each other. This model applies to individuals while they walk, drive cars, fly planes, ride trains etc.
3. The sensor fabric: this concept includes the traditional environment sensor fields as well as the mobile sensor fields (people, car sensor fabrics). This latter scenario is clearly connected with the instrumented constellation scenario

Discussions:
What are the differences of sensor space? How to deal with sensor field? In particular sensors manifest themselves and do integrate operation.
The sensor network is sufficient to do in the network layer.
Possibly doing stuff locally for sensor networks, e.g. environment sensing, temperature sensing. Whether the current network supports the particular requirements?
Do we need further decomposition?
The wireless network is part of the architecture, but is not part of the Internet.
We use constellation to form the subset of the devices.
¡°Dynamic¡±: plug in or out, or mobile.
Single interface or multiple interfaces, having two senses in terms of device and network interface.

Scenario 1.
Mobility tracking:
The point is identity. How to identify devices? Using address? How to naming constellations and devices? How to interpret ¡°address¡±? (It is different from the term used in the Internet)
Why mobile IP failed? What is the requirement for identity?
It requires that the device can get location information from the network.
It is proposed to change the expression ¡°mobility tracking¡± to ¡°ability to route to/from mobiles¡±.
Mobility management is different from mobility tracking.
Is tracking an architecture capability?
One might be able to get the location information from DNS.
It is proposed to add ¡°disconnected operation¡±.
Many discussions about addressing, e.g., how to obtain stable address, how to do delay tolerant delivery (deliver the information to an identity when it plugs in)? Then the network has to provide this awareness.
For geo broadcast, the location is a name.
Questions are brought out: what geo means? How to build the names? Based on geo? Or based on attribution?
So we can combine the identity and geo.
It is proposed to have the flexible definitions of identity and address, to allow route to internet addresses as well as geo addresses and geo regions; multicast and geo multicast.
We need to carefully use the term ¡°address¡±.

Resource is not limited to spectrum.
A case would be: I will help you out if you have limited resource. This is another kind of cooperation.
It is better to be denoted as self-management.
We need to find out what are the specific requirements posed on the Internet.
Is it able to see what is going on in the Internet? Is the network transparent?
The device management is carried out through the network. The management should include the devices, the network elements and the interfaces, not only the user management.
It is good to have the hook to the Internet to find out what is going on in the Internet. But the question is: who is responsible for this?

Security:
Trust is important. It is proposed to add trust to the requirement.
It is also proposed to add privacy to the requirement.
Privacy does not exist in the Internet.
Who should have the right to use some resource such as tracking the location?
Forensics capability is required.
It is also proposed that resource usage should be countable.

Scenario 2.
Constellation:
It is concluded that we use constellation as the term rather than cluster, use opportunistic nets rather than ad hoc networks.
Scope of routing: inter- and intra-constellation.
Whether to use the Internet is dynamic. One can choose routing through the Internet or just do it locally.
Naming scheme is required again. It should not be the same as in the Internet. Probably, it is some combing of the logical name and the physical name.
Thus, the flexible naming is a must.
It is proposed to add naming resolution to scenario 1.

Discover:
Flexible addressing is required again. Consider the case of using RFID to do automatic checking.
Add constellation management.

Multiple outgoing connections:
Aggregating several outing packet to one link is required.

Security:
A question would be: who will certify the operation?
It is hard to use credit to do trust. We probably can use user¡¯s contribution to trace back, like the social trust.
Trust relationship is important. Will the architecture provide trust management?
We need to decentralize the trust management. Then we can overcome the case for example, I want to trust somebody but do not have the right to access the certification.
Trust problem is not solvable in the Internet.
Another question is about information trust. For example, I pass the information about the traffic, will other cars behind me trust me? Is the information true or false?
A possible solution might be some centralized scheme, e.g. the police send out the query and the cars do reply.

Local discovery is a requirement for the architecture.
We can never touch IP, but use data link to do discovery.
Then the car is less interactive with the sensors.
It is proposed to change the expression of ¡°be aware the sensors¡± to ¡°support discover the sensors¡±.

Constellation:
Constellation should be independent of the administration domain.
It is a very hard problem of how to manage the constellation. It is proposed to add membership management.

Scenario 3.
We should make the network as an entity which is invisible.
A question is how about network storage if the network is invisible.
Using overlay.
Delete ¡°dynamic¡±.
Data mining is an example service rather than a requirement. It is proposed to put data mine as an example service.

Since the medium is shared, heterogeneous protocols might exist. Is there any possibility of having other protocols at edge networks?
Add to scenario 2?

QoS:
Is best-effort preferred? Is it a good thing to have QoS support?
Let us do negotiation support.
Does the network need to provide QoS? TCP provides but IP does not.
The network passes the information and allows subset of the network to support congestion control, QoS ect.
Change the expression and add it to scenario 1.

Consider Larry Peterson¡¯s talk: flexibility, reliability, robustness.
Economic incentive as a requirement is added.
We have not talked much about meeting application requirements.

Summary.
1. Routing to flexible identities and addresses
Identities represent devices, constellations of device, geo-areas.
The location awareness is part of the identities.

2. Delay tolerant delivery

3. Visibility of path and topology quality

4. ¡°Identifier based¡± network management

5. Physically decentralized local identifier routing
It is not a global infrastructure.

6. QoS negotiation support mechanism
Constellation management.
What is constellation: set of devices, hierarchy of devices.
Note that devices are treated as the same as the network elements. Constellation management should manage all the devices, network elements and network interfaces.

7. Security/privacy

Questions during discussions:
1. The notion ¡°constellation¡±.
2. It is not the higher layer requirement. What are the more abstract descriptions?
3. How to bind the sensor to the closest AP using address? The term of address has a general meaning, and it should include this capability.
4. Address should not only be IP.
5. What is the minimum requirement for a sensor node to participant a sensor net? Global addressability.