Re: three-tier semantic web application (formerly semantic web tools in a shared hosting environment)

> I'll branch this thread and ask if someone has successfully used a triple 
> store between a Java and a PHP application.

nope, i use my triple store between Ruby, Haskell and Bash applications though. eliminating the need to build DB-engine bindings and SQL-generation for each made this a lot simpler (they all have VFS capability in their standard libraries)

>
> Think about using PHP as a front-end for a SW application which is based on 
> Java. For the front-end I'd like to use Drupal, whose community recently 
> started to experience with RDF also [1] (which is great!!!) Drupal could 
> become a framework for rapid-prototyping of SW applications. However, many 

ive written essentially the upcoming Drupal 7.0 in Ruby, starting 2 years ago and regularly updated/refactored/trimmed since then. the primary test cases were in fact sites ported off Drupal, which at the time did not support RDF (i'll still dislike PHP's syntax and redundancy vis-a-vis ruby/python/llua/perl and unless things change, the spaghetti-party (yeah, it works, but quick refactorability is important)). originally backed by Redland, i had problem's with its query performance, and annoyed by needing a SQL daemon running (and having all the data locked up in a single opaque file only accessible via SQL). so i wrote a new VFS backend. codes at http://repo.or.cz/w/element.git 


most shared hosts provide you with a MySQL database. plus, ive only tested on reiser4, which is extremely optimized for small files. a shared host likely has its FS virtualized at least once, via XEN (and inside that its crappy EXT2 to begin with) then virtualized again via some sort of LVM or RAID setup. so its going to proably suck (i get around 30K triples per second on a fast C synthetic test program, and a few thousand in ruby (MRI 1.9)). as an end user who just needs to store triples and do basic sorted-by-one-attribute queries and doesnt want MySQL churning all the time its great. 


instead of one file per triple, one file per resource could work as well, using Redland to simply (parse into memory model, update, and) write out a new one on edits.. you couldnt query with GNU find anymore though..

> The store should be accessible similarly than RDBMS via JDBC - has there 
> been a discussion about such a middleware protocol yet? Sure, one could use 

afaik Virtuoso has a pretty decent ODBC/JDBC implemenetation. its the de-facto way to use it when theres no native bindings available (there are for Java and PHP)
-c

Received on Friday, 14 March 2008 16:45:38 UTC