Sebastian Trüg wrote:
Hi Ivan,
while this is very good to know it is hard to do it generically. After
all users and developers can create all sorts of queries.
Thanks for the explanation.
Can this be considered a bug in the query optimizer?
Sebastian,
Cost based Query Optimization is a bottomless pit, sadly. This is why
when all is said and done, rules must exist as an option. This isn't
unique to SPARQL, similar issues exist in SQL where optimizers are
overridden etc..
Ivan is articulating an override cost optimizer feature via SPARQL pragmas.
Kingsley
Cheers,
Sebastian
On 05/03/2010 10:20 AM, Ivan Mikhailov wrote:
Hello Sebastian,
With
select * where { ?r a nfo:FileDataObject .
?r nfo:fileName ?f . ?f bif:contains 'breaking' . }
the optimizer decides for some reason that being nfo:FileDataObject is much less frequent
property of ?r than containing "breaking" word in a filename.
That's weird decision because in the EXPLAIN the optimizer says that "breaking" is seldom by
itself, so "breaking" in filename should be even more seldom than "breaking" in anything.
The workaround is to tell the optimizer that you know the order of operations
better than it and to reorder triple patterns:
sparql define sql:select-option "ORDER" select * where {
?r
<http://www.semanticdesktop.org/ontologies/2007/03/22/nfo#fileName> ?f .
?f bif:contains \'breaking\' .
?r a
<http://www.semanticdesktop.org/ontologies/2007/03/22/nfo#FileDataObject> . }
Without define sql:select-option "ORDER", reordering does not help
because the optimizer pay no attention to original order of tables if
there are only few variants. (If the query is complicated and no define
sql:select-option "ORDER" is specified then writing patterns in some
specific order guarantees that the optimizer will consider this order of
execution among other possibilities even if there's not enough time to
search throughout all possible permutations.)
Best Regards,
Ivan Mikhailov
OpenLink Software
http://virtuoso.openlinksw.com
On Fri, 2010-04-30 at 15:10 +0200, Sebastian Trüg wrote:
Hello Hugh,
Please find the output attached to prevent line breaks.
Cheers,
Sebastian
On 04/30/2010 03:14 AM, Hugh Williams wrote:
Hi Sebastian,
Can you please provide the explain output detailing the execution plan for both
queries for comparison, as detailed at:
http://docs.openlinksw.com/virtuoso/perfdiag.html#perfdiagqueryplans
http://docs.openlinksw.com/virtuoso/fn_explain.html
Note you must include the "sparql" keyword before the start of each query.
Please also provide the output of running the status(); command to provide access
statistics on the server.
Best Regards
Hugh Williams
Professional Services
OpenLink Software
Web: http://www.openlinksw.com
Support: http://support.openlinksw.com
Forums: http://boards.openlinksw.com/support
Twitter: http://twitter.com/OpenLink
On 29 Apr 2010, at 16:40, Sebastian Trüg wrote:
Hello,
using a V6 server with default indexes I would like to know why the
following queries are different in performance. The store contains
thousands of different graphs but adding a "graph ?g {}" around all
patterns does not change the execution time.
select * where { ?r a nfo:FileDataObject .
?r nfo:fileName ?f .
?f bif:contains 'breaking' . }
takes a long time to finish - almost a minute.
select * where { ?r nfo:fileName ?f .
?f bif:contains 'breaking' . }
is finished in no time.
I just would like to understand the reasons behind the difference. If
possible also a solution. :)
Cheers,
Sebastian
------------------------------------------------------------------------------
_______________________________________________
Virtuoso-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/virtuoso-users
------------------------------------------------------------------------------
_______________________________________________
Virtuoso-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/virtuoso-users
------------------------------------------------------------------------------
_______________________________________________
Virtuoso-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/virtuoso-users
--
Regards,
Kingsley Idehen
President & CEO
OpenLink Software
Web: http://www.openlinksw.com
Weblog: http://www.openlinksw.com/blog/~kidehen
Twitter/Identi.ca: kidehen
