Project Page for CS 294: Deeply Embedded Networks
Rabin K Patra & Sergiu Nedevschi {rkpatra,sergiu}@cs.berkeley.edu
However it is concievable that in some situations this may not be a valid assumption. For example sensor/actuator networks deployed over a wide area might have to deal with environments where a continuous end-to-end path cannot be gauranteed. Example systems include the Zebranet([ZebraNet]) that used peer-to-peer networking techniques in a mobile sensor network designed to support wildlife tracking A pertinent example is the sensor node deployment at the Great Duck Island([GDI]) that reportedly suffered from loss of data when the base station failed - this could have be avoided if the sensor nodes had queued the data until it was safe in persistent storage.
Also, small low wireless range personal handheld devices that have applications in social networking and providing ICT in developing regions also have to deal with intermittent connectivity.
Delay Tolerant Networking(DTN) is a new field that looks the architectural and protocol design principles([DTN]) arising from the need to provide communications with and among extreme and performance-challenged environments where continuous end-to-end connectivity cannot be assumed. An API has also been proposed that provides different classes of traffic to end devices so that applications can be easily developed in DTN environments.
Though DTN has been implemented for the PC platform, we want to explore the issues in adapting the protocol to a TinyOS like platform . Also an event-based system like TinyOS could probablu be the preferred choice for future power and memory constrained devices.
The DTN architecture currently consists of regions and DTN gateway nodes that use store and forward for messages. It also includes concepts of hop-to-hop custody transfer for reliability.
Our first step is to analyse the requirements of such a layer in a sensor network environmentand determine what of the features need to be supported. The main challenges in implementing such an API on TinyOS like platforms would be the smaller storage,memory and CPU resources. Thus we need to look at a lightweight version of the protocol that can function with limited persistent storage and computation power. At the same time the asynchronous nature of message delivery in a DTN lends itself naturally to the TinyOS model of execution.
A basic design would have the sensor network attached to a gateway(or proxy) with persistent storage. The sensor nodes would queue their data until the proxy takes custody of their data. The DTN Lite layer will also ensure that the applications get the correct indications from the network regarding the status of their messages.
Another particular goal for this project is to achieve interoperability between the wireless sensor networks running DTN Lite and and the Internet running DTN.
Initially, we intend to implement and simulate the functioning of DTN Lite using TOSSIM. This step will enable us to estimate the complexity of the design as well as the amount of memory and computation required. Depending on the results of this step, we may attempt a physical deployment on MICA motes.
We intend to evaluate our design using a simple application, namely the message transfer from one end to another in a sparse sensor patch. Some other applications suitable for our design include data collection and delayed queries for multi-tier sensor patches that can only be accessed occasionally by data mules, as well as any other application running in a mobile and intermittent sensor environment such as ZebraNet.