RE: RDF DB Shootout: Preliminary Results

From: Seaborne, Andy <andy.seaborne_at_hp.com>
Date: Wed, 11 Jan 2006 13:39:27 -0000

-------- Original Message --------
> From: Danny Ayers <mailto:danny.ayers_at_gmail.com>
> Date: 9 January 2006 17:46
>
> On 1/9/06, Dean Allemang <dino_at_happydino.com> wrote:
>
> I'll see if I can track that down, and get you some
> > answers (if I find them, do they go to you directly or to the list?)
>
> I think such info would be welcome on the semanticweb_at_w3.org list,
also
> added to/linked from the ESW Wiki, somewhere around:
>
> http://esw.w3.org/topic/LargeTripleStores
>
> I'm eagerly awaiting some response to the shootout from the Jena folks
> ;-)

Looking at the tests, there seems to be no transaction control for the
Jena tests. With JDBC, each operation will get an implicit transaction
wrapper. This is bad for read access and very bad for update.

  public void startTransaction(/*=RepositoryType=*/Object/**/ repo) {}

I don't have experience of using Derby or its JDBC driver but it does
appear to support transactions so maybe it isn't a no-op. As triples
are small, the transaction overhead can be more than the operation
itself.

Also

        public void commitTransaction(/*=RepositoryType=*/Object/**/
repo) {
        model.close();
        model = null;
        try { conn.close(); } catch (SQLException e) {}
        conn = null;
        getModel();
        }

opening and closing models will achieve transaction effects but it's
expensive. You shouldn't need to open and close connections - if there
is some reason why this needs to be forced for Derby, I'd be interested
in hearing about it.

It says
   tester.weakAPI = true;

and if weakAPI is true, it does:
      tstDriver.commitTransaction(repo);
      tstDriver.startTransaction(repo);
every 100th time.


Sorry if I've misread the code or my lack of Derby knowledge has missed
something obvious.

Vineet - we talked about this offlist - let me know when you less busy
and we test this and may be a direct connection to Derby.

- - - - - - - - - - - - -

More broadly:

If anyone is interested in collecting datasets and SPARQL queries
together so that users can do SPARQL performance testing without having
to spend so much time on their test harness, then let's talk about how
we can get this to happen for the community.

        Andy

>
> Cheers,
> Danny.
>
> --
>
> http://dannyayers.com
Received on Wed Jan 11 2006 - 13:39:21 EST

This archive was generated by hypermail 2.3.0 : Thu Aug 09 2012 - 16:39:18 EDT