Urban sensing, community (wireless) networks, intelligent edges & social tagging (Part V of X)

We still haven’t talked about social tagging, community networks and urban sensing. We will now !

First I have attached my presentation for my talk “Urban Sensing, the place in-between work and play 2010 & Olympics 2008” at CINACON 2006 (Chinese Information & Networking Association) . I will explain the presentation as a starting point and then continue with a discussion on our exciting collaboration with UCLA REMAP/CENS). The project is just starting, and I will report the progress time to time …

Urban sensing, community (wireless) networks, intelligent edges & social tagging (Part IV or X)

Wireless AP feature set for an Intelligent Edge for a flexible sensor information fabric

Now, we come back full circle, idea to social dimension to architecture back to technology! From a technological capability point-of-view, we need a few well chosen “intelligent edge” primitives embedded in a device.

In a wireless community network, the AP is the best choice; or an embedded system which hangs off of the AP would suffice.

Let us look at them in random order ;o)
a)An Intelligent Edge should participate as a first order sensor network device
While a gateway could pass as a conduit, it will not be able to understand, and act as a control/management plane for a sensor network. This is essential when we are looking at “native” interfacing between the internet and the sensor network fabric. As we discussed earlier, an intelligent edge need to perform functions like sensor clustering, data-centric routing and Local collaborative in-network information processing.

Of all the feature sets this is one of the toughest ones – the device needs to talk many wireless protocols like 802.11, 802.15.4, 802.16, … It also needs to understand Layer 2 and Layer 3 variations of the above and be able to converse the static and mobile network protocols with all the security primitives thereof ! I am hoping for a single chip with all of the above! Of course, seriously, we might not be able to get a single chip, but a set of chips that address groups of the functionality and then we can mix and match. Also we could implement some parts, either in firmware or as software overlays.
And we need to get all in a $100-range for consumer-class and as a part of bigger offering for the enterprise-class and carrier-class devices.

b)An Intelligent Edge should be capable of injecting context information into the fabric
The figure 1 in CENS Research project TENET (http://research.cens.ucla.edu/projects/2006/Systems/Tenet/default.htm) depicts the intelligent edge concept. The Urban sensing project from REMAP http://research.cens.ucla.edu/projects/2006/Systems/Urban_Sensing/ has some very specific requirements for an Intelligent Edge.
b.1) Location & time
The two pieces that an edge device can authoritatively declare are location and time ! As the Urban Sensing proposal eloquently says “location and time are key elements of context that can be attested to by the network itself. Essential to building space-time semantics into the network fabric is the ability for the network to verifiably measure location and time of a device when it injects a packet into the network”
b.2) Cloud functionalities – discovery, naming, resolving end point location
There will be thousands of sensors distributed over a geographical position and it is impractical to visit each one of then for configuration and command control. Moreover, if we want to address the sensors for policy enforcement, we need naming, discovery and related primitives.
c)An Intelligent Edge should be able to act as an edge policy intermediary
A policy intermediary is a big word, with lots of nuances – even the word policy will bring an army of folks with different interpretations … From this perspective, I defer to my esteemed colleague John who says :
“The key is packet/message classification i.e. be able get info all the way from MAC address and VLAN to SOAP Header/Xpath. Provide the ability to use the classification by a declarative policy framework that works with the message classification (abstracted such that new protocols can be seamlessly added), with an interface to allow applications to inject policy into the network. Add an extensible services framework to allow the addition of execution elements to the policy ( the E in PEP) and we are home free.”
d)An Intelligent Edge should be able to implement policy based context derived privacy
A major concern of any sensor network information fabric is privacy. An essential feature is the resolution control of data publishing, enabling selective sharing of information. Quoting from the Urban Sensing project “the edges of the proposed architecture must be capable of deliberately reducing the fidelity of the context information it measures (location, time) or derives from sensor values, and it must do so differently for different subscribers.
e)An Intelligent Edge should perform all these functions with primitives native to sensor networks – including power conservation, power awareness, optimum verbosity, Self-forming and self-organizing, self-healing, …
f)An Intelligent Edge should be an orthogonally extensible platform
Finally, we cannot achieve #a) thru #e) above with out the flexibility and capability to add things over a platform. It definitely is not an unbounded server platform but a facility to add behaviors either influenced by policies or activated by some form of scripts. Any involved functionality would need to be added outside the platform based on some network protocols and leveraging existing capabilities. Again referring to our esteemed colleague John “Focus this on open protocols and allow new protocols to be added using a descriptive language. With the basics you can start addressing most problems – the key is to get the basics right and then build on it with application partners in both industry and academia.”

• I think we have discussed the essential fundamentals of what an intelligent edge would like in many dimensions and how a wireless AP can be one.
• Intelligent Edge = Network protocols + Algorithmics + Context + Declarative Policies
• Thoughts ?
• Insights ?
• Counter-arguments ?

Urban sensing, community (wireless) networks, intelligent edges & social tagging (Part III or X)

Architectural Elements of the Intelligent Edge Networks

Originally I wrote parts of this blog from Paris – while attending the Ubuntu Summit. An interesting set of activities, I will have to write about it one of these days …

For now, let us concentrate on the architectural elements that have been discussed conceptually in various research papers as well as pragmatically from various projects …

A) Sensor Network stack has layers similar to the OSI layers
An excellent paper from UCB [1] describes a sensor network in terms of :

• L3: Packet routing,
• L4: Local collaborative information processing,
• L5: Wide-Area data dissemination,
• L6: Wide-area collaborative information processing and
• L7: User-level tasking and querying.

In this model, the L3-L5 are Network Level Events(NLE) which should be handled by an edge platform say a wireless AP. The L6 and L7 are definitely application responsibilities.

B) But unlike the internet, routing is an overhead in a sensor network
One important characteristic of the sensor networks, that separate them from the internet, is the fact that a sensor network predominantly derives utility from the data it gathers, while Internet gathers most of its utility from routing the data.

C) And so a data-centric routing approach makes more sense
The identity of node is less important i.e. the content is more relevant than the node itself. In short, what we need is data-centric routing – optimizing the path to a piece of information (or from a piece of information) with the minimum power!

D) So does sensor clustering …
Many sensors in the same vicinity will have redundant data and for optimization all of them need not transmit the data across a whole sensor network; an intelligent edge can instruct the sensors when to transmit and when not to transmit, thus optimizing the network for redundancy, resiliency and accuracy while conserving power and bandwidth! This requires semantic capability (i.e. the edge needs to know what the data is all about and where they come from) with context based information …

[1] Data-Centric Storage in Sensornets with GHT, A Geographic Hash Table

Urban sensing, community (wireless) networks, intelligent edges & social tagging (Part II or X)

Social Dimensions of the Intelligent Edge Networks

Let me dive into the social dimensions of the intelligent edge, as a prelude to my favorite topic – The Technology Dimensions!

There are many compelling reasons to believe that sensor networks will be a big part of our lives – at work and at play, more importantly the places in between; my focus, in some sense, is the in-between place which the Urban Sensing is what all about …

A) Pervasive sensor networks are an eventual necessity and hence a reality

Life is getting more complex – even mundane things like traffic are not predictable and takes a lot of our time. I read in [1] that Warren Weaver in his address to the Rockefeller Foundation suggested problems are of three kinds – problems of simplicity, problems of organized complexity and problems of disorganized complexity. Jane Jacobs observed that cities have progressed from problems of simplicity to problems of organized complexity. We really do not have the time nor the processing cycles to manage the organized complexity for our daily lives; moreover the complexities will increase exponentially … Hence, we would be better off delegating some of the mundane, yet essential complexities to the machines …

B) But we need better interaction interfaces …
Currently the way we interact with the sensor environment is very archaic ! Looking at it holistically, we know they are there; they know we are there; but do they know that we know that they are there ? This is the principle behind many feedback systems like the tactile feedback in the keyboards, the blinking lights on the car alarm systems… they are all for our benefit not theirs !

C) Which means, the sensor network should be syntactically visible and semantically transparent
We do want to see them, be able to change the way they behave, we also need more selectivity in what they see and what they transmit … But we do not care if they use 802.15.4 or ZigBee or ZeroConf or all of them . We, as technicians care, but as users of these technologies, we do not need the visibility to that depth. But we do need a wireless dash board that shows the signal strength and if the networks we are going to connect are secure …

D) But this extreme edge networks cannot be internalized by the current Access-Distribution-Core model of the traditional networks …

The Core focuses on speeds and feeds, the fastest transport of the packets and defines the diameter of the network; The Distribution is about policy based connectivity and the Access is all about work group connectivity.
All the three assumes a flow from servers to the edge; but the flow in sensor networks are asymmetric (is lots of small messages flow from the edge to the core !) and of different characteristic (lots and lots of small messages than larger messages)

E) And Privacy and context based declarative policies are very important; they need to be built-in the network substrate – let me repeat privacy and policies are neither optional nor be an after thought
Like I said earlier, we need to control what flows in and out of a sensor network, at the edges ! In this regard I had some interesting conversations with our engineers; with UCLA’s Jeff Burke and Deborah Estrin as well as with folks from Monash university in Australia, Stanford University, UCSD, … and of course with our customers who are deploying sensor networks in their assembly plants, integrating with their logistics systems, deploying first responder and hastily formed networks et al …

We will discuss the Technology dimensions and challenges next …

[References]

[1] Cities and Complexity By Michael Batty

Urban sensing, community (wireless) networks, intelligent edges & social tagging (Part I or X)

I have been wanting to write about this for some time. I did write a few blogs somewhere else which is no more on the face of the earth ;o( So let me restart …

Intelligent Edges …

I am not a fan of application-gateways in an architecture; while I think they are ok as a patch-work and a necessary evil in an operational setting ,conceptually they are terrible. I am more tolerant of protocol gateways, though they should also be replaced by native protocols in due course of time.

Prelude

Let me explain …. The story began an year ago, when I started working with sensor networks … The network substrate in sensor networks has slightly different characteristics (than normal networks) mainly due to power and resource constraints and usually the interface to the mother-ship is a gateway – just throw all the data over the wall to an IP network … and I think this is a bad design and what we need are intelligent edges at the sensor network borders …

An exciting summit and a few thoughts …

Few weeks ago I was fortunate enough to be invited to the Urban Sensing Summit at UCLA by Deborah Estrin and Jeff Burke. I was also on a panel with IBM’s Rod Smith, Spelling TV’s Ken Miller and Disney’s Bruce Vaughn. The panel was around user experience, intelligent inferences based on context and many of the edge functionalities. In the afternoon, I attended a break out organized with Jim Mutter, President, Premier’s Technology Council, British Columbia and Jason Brush, Senior Vice President, User Experience, Schematic. The talk was around Vancouver and the Olympics 2010 and what we could do around urban sensing – again intelligent edges, making inferences out of the myriad of data coming from a multitude of sensors …
Coming back to the first paragraph, while I am not sure what the complete solution architecture would look like, I know what it will not contain – data/application gateways !

Intelligent Edges for a Sensor Network …

Synthesizing lots of research in this domain, the functions required for a sensor network edge are –

1. An intelligent edge (Sensor clustering and overlapping, context aggregation),
2. A control and management plane (network scalability, redundancy, accuracy and resiliency),
3. A policy mediator and distribution point,
4. A policy evaluation point (PEP) and a policy execution point (PXP) and
5. A security materials distribution intermediary.

We will dig deeper, into a few of these ideas, including the various functions that are required from a substrate/platform, in the next few blog entries …

Streaming Video Network over 802.11 wireless

We are building a streaming video for the North CA First Lego League tournamnet. Will blog the efforts, the choices made and hopefully the successful results (the tournament is in San Jose City College on 20 Jan,06.

Challenges: (I will write notes on each component/sub system in the configuration section)

Range – Trying to solve it by 2 X 400mW 802.11g cards at the AP and 200mX USB client cards! The AP is MIT roofnet Sorkis board with 128 MB.

The gym, where the main event is, might be a big Faraday cage ! Still need to do a site survey …

Video server – videoLAN on a Shuttle SN27P2 is one choice. We have two streams plus the score sheet. So need to aggregate them in some way. (TBD as on Dec 17,06)

Configuration

Shuttle XPC SN27P2 Configurations

As a part of a video streaming network I am building for the North CA First Lego League Robitics competition, I assembled a shuttle PC. Will blog the configuration and the logic behind the method behind the madness:

I bought them from directron, they have good prices, quick turnaround and good service.

Before we jump into configuration, some lessons learned:

1. Ask them to assemble it. Costs only $29 and takes lots of hassle. But OTOH, if you get the parts and assemble them yourself, you get the experience (as I did). My 8 year old got a kick out of assembling the CPU, fan et al. But until it all worked, I was wondering if I had ruined any connectors or chips.
2. For this suttle case, get the IDE optical drive and use the SATA for hard disks
3. This does not have PS/2 keyboard or mouse. I have the SBT-PS2U adaptor which takes a PS/2 keyboard and a mouse and converts them to USB. Works fine. I can still use my favourite Microsoft keyboard !

Configuration notes:

• Shuttle Case – SN27P2

• CPU : AMD Athlon 64 X2 3800+

• No clue why the +. Is there a 3800 – or do they plan to have a 3800# ? ;o)

• Optical drive : Plexor PX-755SA SATA 16X

• This was a mistake. Becuase the case has 2 SATA cables and one IDE connector. To have a RAID 5 we need 3 disks and the Optical drives takes one SATA.
  • [12/17/06] I am in the process of ordering an IDE optical drive and 2 more Barracudas for a good RAID system. Naturally you do not want your system to fail during a real time video streaming session! Will add notes how it went
  • Hard Disk : Seagate Barracuda 7200/320 GB HDD ST3320620AS
  • Good price-performance. I will have 3 of these in a RAID 5 combination.
• RAM – Two kingston Value RAM 1 GB each, DDR2-667
  • I didn’t realize but these are 4 X 512 MB chips. But as the SN27P2 has 4 slots, it is OK. Again, price-peformance consideration.
• Display : PCI Express Sapphire Radeon X1300XT/512 MB
  • Again a good card with 512 MB, still within budget.

Fedora on Dell E1505/Core Duo 2 7400/ATI X1400

Ubuntu works reasonably, but I wanted to see how Fedora Core 6 performs. Here are the agonies and ecstasies !

Step 1: install Fedora Core 6

Worked fine. It recognized the ATI X1400 and so was able to install, unlike Ubuntu which hung up. But the screen resolution is 1400 X 1050 instead of 1680 X 1050. Possibly need to download the drivers fron ATI.

The Fedora core 6 recognized the dual core CPU.

Step 2: Wired Networking

Networking. Interestingly the network is not working. Need to figure out why.

Step 3 : Wireless Networking

The intel wireless drivers are at http://downloadfinder.intel.com/scripts-df-external/filter_results.aspx?strTypes=all&ProductID=2259&OSFullName=Linux*〈=eng&strOSs=39&submit=Go%21.