- From: Dave Reynolds <der@hplb.hpl.hp.com>
- Date: Thu, 23 Aug 2001 11:05:31 +0100
- To: Dan Brickley <danbri@w3.org>
- CC: www-rdf-interest@w3.org, Mark Baker <distobj@acm.org>
Hi Dan, > I've recently been experimenting with SOAP interfaces for RDF query > protocols; while it works (the Python and Perl SOAP libraries made things > quite easy for me as a developer), I have some worries about the way SOAP > uses HTTP (eg. all my queries get represented as HTTP POSTs, where the > cacheable GET would make more sense). The RESTwiki seems like a good place > to record these experiments; I hope to make time to do that in the near > future. I've also been experimenting with building a SOAP interface to an RDF store. In my case this was in Java using the Jena RDF toolkit and the Apache SOAP 2-2 toolkit running on Tomcat. Whilst it worked in the sense that I can create, query and navigate RDF models in the remote store I did find performance to be a problem. I ended up using SOAP-with-attachments and transferring the bulk RDF data as a MIME attachment in N-triple format and just using SOAPencoding for specifying the operation required and the actual parameters (e.g. query specification). Even then the overhead of going via SOAP was large - in typical configurations I was seeing around 130ms latencies due to a mixture of the overheads of XML parsing of the SOAP packet and problems with getting Apache SOAP to stop using TCP slow-start. I believe I can get this down to more like 40ms but that compares badly with simple HTTP GETs on a servlet with arguments URL-encoded which takes under 10ms on the same network/machine configuration and data volumes. This left me feeling that a simple direct HTTP usage pattern for remote RDF query might be a useful alternative to a full SOAP interface. I'll be very interested to hear how your experiments develop - keep us updated. Cheers, Dave
Received on Thursday, 23 August 2001 06:06:36 UTC