Wednesday, May 19, 2004

Semantics in P2P and Grids Notes

Not having looked deeply into this in the past except at the The Semantic Grid web site I came into the workshop without much context as to what the talks were going to be about.

The first talk was about using query expansion on a peer to peer network. It extended Limewire to allow different nodes to expand the keywords using something called Keyword Relationship Databases (KRDB). Weightings are graded based on how successful the data they return is.

The second speakers presentation was about using Small World Networks (SWN) as a way to scale P2P networks. You place the data and the nodes in a k-dimensional space. The problem then is that you have to keep the number of dimensions reasonable. The speaker gave an example of flattening a 2-dimensional space into a 1-dimensional space by basically turning it into a linked list.

Bibster was the topic of the third paper. This used a combination of SeRQL, Sesame and JXTA to implement a distributed bibliographic system. It also used similarities in the metadata to route queries and remove duplicates.

The next talk focused on creating truely distributed P2P databases. It appeared that they were using something that looked liked lattices to implement schema matching between peers. This was called "correspondence rules". There was also "coordination rules" which dictacted how queries were forwarded to peers. It also used JXTA. The main feedback seemed to be that the paper gave names to these problems and but didn't really solve them.

coDB was the next topic of discussion. Which seemed to be a way of doing First Order Logic (FOL) over P2P networks. I had a problem understanding the point of this one. For example, the paper mentioned that if peers that created the result drop out then the answers from that peer should be removed. The feedback from someone else was whether it was useful in an open network to do this. The example brought up, was marriage. He was saying a peer would make the inference that if one person in a marriage was male the other would be female. And that brought up the rather topical example of gay marriages. What one peer inferred maybe different for different reasons than another peer. This was a bit beyond the scope of the paper though it seemed like a valid criticism.

The sixth talk was done by a someone from HP. This was pretty much the highlight of the morning. Because it specifically talked about RDF and OWL and they used Jena (of course). The main thing to come out of it is that they wanted to be able to calculate properties ("virtual properties") expressable in RDF/OWL. For example, server price is a funtion of software and hardware price. They created a type of TransitiveProperty called FunctionalProperty. A functional property would define a function like "foo = x * y + z". How this would be defined in OWL hasn't been defined as far as I know. It sounds similar to aggregate functions like average, sum and count but are fully configurable. I vaguely remember someone at work talking about this. Obviously, a cool idea anyway.

The last three talks were about trust, GridDB and ebXML. They were fairly short but interesting. Probably the most interesting thing I got out of it was that OASIS (who are working on ebXML) are trying to be inclusive of current technologies like Topic Maps and RDF/OWL. In fact, the speaker did say that UML was going to be expressable in RDF/OWL. I also thought that GridDB looked a lot like our current TMex product.

Anyway, apart from that I'm really enjoying the conference. Google set up a booth to try and get people to work for them - little did they know I was only after their pen.

No comments: