A. Sundararajan's Weblog

This talk is about pervasive computing (a.k.a ubiquitous computing) with products from Sentilla. There was an interesting demo about humidity sensor detecting changes and sending a message to a host. The "motes" run CLDC 1.1 VM (+ proprietary profile for motes). These motes have ports for sensors and actuators and some built-in sensor. There were many interesting suggestions for embedded programming for such small devices (don't allocate in inner loops and there by leading to to GC kick-in, avoid too many static fields, avoid threads whenever possible and so on).

The basic idea is to run JavaME applications (developed for different configurations/profiles/subsets of optional packages) on top of JavaSE. The extended JavaSE classes and packages not available in specific profile or optional package set [implemented by a specific phone] should not be made available to JavaME apps targeted. i.e., only the classes available to a specific phone model should be available. If the JavaME app tries to access any other class, it should receive ClassNotFoundException. The speakers explained how to achieve such "containers" by class loader based isolation. The problem is that they seem to solve only the class access. What about extended methods and fields? For example, platform core classes on JavaSE have superset of methods [more methods on the same class available on JavaME - eg. java.util.Hashtable has more methods on JavaSE). The application classes have to bytecode analyzed and instrumented to take care of field/method accces. It seems that their current product that does not address this yet.

The discussion started with some nice demos. There was a demo with AllegroGraph RDF store, Twine, a demo with using GRDDL and getting RDF triples by a proxy server. i.e., a proxy serves does the GRDDL transformations to get RDF triples from sites [which could be stored/analyzed with RDF stores subsequently] and a demo with FOAF files. Interesting take aways from the discussion include:

• We don't have to wait for SEMANTIC WEB with full fledged reasoners and so on. Instead, add little semantic bits to existing web (say using RDFa, GRDDL etc.) in your current web projects/pages.
• There are many Java tools. There is need to standard Java APIs for triple store access etc. Right now, we have to write for Jena, Sesame etc. It was also felt that APIs will need to wait for more usage scenarios.
• There are tools to expose your existing databases as virtual RDF stores -- for example: D2RQ. Probably, most of the RDF triples could come from existing data.
• Privacy, security of the information is very important. Work needs to be done in this area.
• Natural language processing and getting triples out of it is very hard. You may want to refer to systems like DBpedia.