RDB2RDF Working Group Teleconference -- 26 Jan 2010

<trackbot> Date: 26 January 2010

<mhausenblas> scribenick: whalb

<mhausenblas> cygri is with mhausenblas

roll call

approve minutes

http://www.w3.org/2010/01/19-RDB2RDF-minutes.html

<mhausenblas> +1

<Ahmed> +1

<mhausenblas> ACTION-18?

<trackbot> ACTION-18 -- Harry Halpin to update web-page and schedule -- due 2010-01-19 -- OPEN

<trackbot> http://www.w3.org/2001/sw/rdb2rdf/track/actions/18

<mhausenblas> close ACTION-18

<trackbot> ACTION-18 Update web-page and schedule closed

<mhausenblas> ACTION-19?

<trackbot> ACTION-19 -- Harry Halpin to put Souri's draft into a WBS poll and notify the WG -- due 2010-01-19 -- OPEN

<trackbot> http://www.w3.org/2001/sw/rdb2rdf/track/actions/19

<hhalpin> will ping souri re e-mail.

<hhalpin> which part of the draft in particular?

<Souri> We have put a write-up (including an XML syntax) on the wiki

<hhalpin> send me the URI

<hhalpin> (looking)

<Souri> under Working docs

<mhausenblas> ACTION-20?

<trackbot> ACTION-20 -- Eric Prud'hommeaux to update http://www.w3.org/2001/sw/rdb2rdf/wiki/LinkedDataAspects#Linked_Data_Principles to say 'MUST allow to..." -- due 2010-01-26 -- OPEN

<trackbot> http://www.w3.org/2001/sw/rdb2rdf/track/actions/20

<mhausenblas> close ACTION-20

<trackbot> ACTION-20 Update http://www.w3.org/2001/sw/rdb2rdf/wiki/LinkedDataAspects#Linked_Data_Principles to say 'MUST allow to..." closed

<mhausenblas> ACTION-21

<mhausenblas> close ACTION-21

<trackbot> ACTION-21 Update http://www.w3.org/2001/sw/rdb2rdf/wiki/LinkedDataAspects#Well-known_Entities to reflect discussion (3 cases) closed

<mhausenblas> ACTION-22?

<trackbot> ACTION-22 -- Eric Prud'hommeaux to add Ahmed's plan to the Wiki -- due 2010-01-26 -- OPEN

<trackbot> http://www.w3.org/2001/sw/rdb2rdf/track/actions/22

<mhausenblas> close ACTION-22

<trackbot> ACTION-22 Add Ahmed's plan to the Wiki closed

<mhausenblas> ACTION-23

<mhausenblas> ACTION-23?

<trackbot> ACTION-23 -- Richard Cyganiak to update http://www.w3.org/2001/sw/rdb2rdf/wiki/LinkedDataAspects#Provenance re today's discussion -- due 2010-01-26 -- OPEN

<trackbot> http://www.w3.org/2001/sw/rdb2rdf/track/actions/23

<Seema> Souri I will follow up with you on this

ahmed: main topic is how we continue forward, give an outline and set direction

<hhalpin> sounds good to me

R2RML Mapping aspects

ahmed: two sub-teams, another team for linked data needed

<hhalpin> well, sort of.

soeren: no further linked data group needed, included in both other sub-teams

ahmed: wants linked data area to be covered anyway
... need more people commited to sub-teams
... how are we going to proceed?

<hhalpin> +1 status reports

<Souri> This was the URL: http://www.w3.org/2001/sw/rdb2rdf/wiki/RDB-to-RDF_mapping:_A_SQL-based_Approach

ahmed: are teams going to give status reports of e.g. 15mins?
... how are we going to show progress?

<hhalpin> http://neurocommons.org/page/Common_Naming_Project

<hhalpin> http://www.okkam.org/

<hhalpin> http://linkeddata.org/

hhalpin: suggest that each group has a member dealing with the different aspects of linked data, etc.

<hhalpin> i'd also suggest that we have a session with the HCLS group

mhausenblas: we should focus on mapping stuff in the first phase, other members supporting both teams re. linked data

<hhalpin> as they have interaction for NeuroCommons Common Naming Project

mhausenblas: whenever there is a need from one of the groups to get some input on linked data topics the experts should chime in

<hhalpin> ericP?

<hhalpin> He's the staff contact for HCLS

hhalpin: also input from the life science domain, probalby from ericp

<mhausenblas> ACTION: ericP to take care of the HCLS statement [recorded in http://www.w3.org/2010/01/26-RDB2RDF-minutes.html#action01]

<trackbot> Sorry, couldn't find user - ericP

<hhalpin> hmm

<hhalpin> let me look

soeren: two proposals: oracle created one and soeren tried to distill another one, wanted to align efforts
... not yet talked to each of the other groups
... suggestions for possible syntax of the mapping:

<soeren> http://www.w3.org/2001/sw/rdb2rdf/wiki/R2RML_SQL_View_Annotation

soeren: called SQL View Annotation, intention is to add syntax for mapping language
... syntax: db conn needs to be established, namespaces + prefixes
... two distinguished namespaces: base namespace for creating data instance items
... optional schema namespace: properties, classes
... could be overwritten by custom definitions
... table definition is the core, most important part
... R2RMap current single syntactical element, first parameter basically a relation identifier
... second element: path used to create linked data entity identifiers
... additional parameters are optional
... third element: reference to a class (rdfs, owl) used to make instances of this specific clas (optional parameter)
... remaining parameters: definitions for each of the columns in the relation (from param1)
... they are all optional
... if omitted the column name is used for creating identifiers
... syntax is inspired by triplify
... additional issues: user defined output processing functions, static metadata, attribute-value tables, named graphs,
... update logs, provenance (triplify approach could also be applied here)
... a lot of work already done on how to tackle these issues
... core is r2rmap syntax
... as a staring point that can be refined later on

<Zakim> mhausenblas, you wanted to note on Test Cases

mhausenblas: start thinking about test cases: specify sample input and output
... ideally a set of test cases is created/made available

ahmed: alignment with ashok?

soeren: contacted ashok/oracle, waiting for responses
... discussion will take 2-3 more weeks

<Souri> Soren, not sure if you got our email from Friday -- we wanted to find out times that would be convenient to you

ahmed: two teams working together, is there a need for a leader of the sub-teams?
... responsible for communication, coordination, not a technical leader

mhausenblas: person in charge should also have reporting responsibility

ahmed: any volunteers or suggestions?

orri: we should do something that substantially relies on sql
... fine with something that's compatible to sql views as long as it works out in the end

<Souri> Did anyone get a chance to review the XML syntax we had posted last week? (special case simplification with auto-mapping is always possible)

<hhalpin> For soundness and completeness, I would defer to Marcelo or at least have him go over the draft :)

<Marcelo> sure, I can do it :-)

<hhalpin> ericP: I think it's left stuff to specify mapping SPARQL to SQL

<mhausenblas> scribenick: hhalpin

soeren: but SPARQL constructs are very limited in expressivity
... assigning results of a computation to a new variable
... build a string by concating 2 others
... i.e. firstname+lastname concat for a URI

<cygri> hhalpin, its richard

ericP: can't do that.

cygri: And we are not sure about SPARQL 1.1
... we will have use-cases outside of SPARQL construct
... this will require a bit of work
... if it is possible

ericP: understood

orri: it would seem to me
... that we need to things in sql
... but we should take sparql 1.1 as a given

ericP: but the point remains
... that a bit of sql
... the commonality is still more expressive than even sparql 1.1
... what do we save about this use-cases?

Ahmed: the main issue with SQL view
... is you have to "touch" the database

<whalb> soeren: disagrees

<whalb> ... no need to touch the database

<Souri> I disagree

<whalb> cygri: several database problems with horrible performance

<Souri> agree with Soren

<whalb> soeren: customised mapping approach still could be done, optimised subqueries will be faster

<whalb> orri: sql dbs will be fine

<whalb> ... query gets unusual when mapping sparql to sql

<whalb> ... in prinicple should be fine

<mhausenblas> scribenick: whalb

cygri: sparql to sql translation is hard for the user to optimise

Dan: manipulation can be done on internal representations

cygri: you can use base tables if that is all you need

<juansequeda> yes

cygri: need to rely on optimiser in database if complex query
... not same effort as doing optimisation on other side

ahmed: subqueries not a problem
... problem is at the end you get sql datatypes

cygri: not a problem, soeren's syntax proposal covers that basically
... can be easily done outside db

ahmed: why classify this approach as sql view
... translating sql data outside the db

soeren: there is need for a mapping engine finally
... sql (view) annotation approach
... sql engine already does mapping from sql data to data

ahmed: disagrees, you get back sql datatypes if mapping is done outside
... why call this approach sql view mapping

cygri: sql output is turned into rdf triples
... transforming to sql view is done inside db

<ericP> SPARQL queries tranlated to SQL

<Souri> The SQLdefString is the one that is employed directly against the host db (i.e., the data source)

cygri: transformation from three col sql result is simple work done outside
... mapping is inside the db

ahmed: view is defined term in sql world
... would not call that approach sql view, misuse of terminology

<hhalpin> We can call it something else I imagine...

orri: column is the predicate on the row, multi-col views ok

ahmed: sql view means that data is updated automatically
... the last mentioned approach does not need any special optimisation

cygri: still a problem for specific db systems, e.g. mysql
... simple looking queries can perform poorly

<mhausenblas> we need to wrap up soon, we're on the top of the hour

ahmed: which subqueries do not perfom well?
... mysql usually only has drawbacks in transactions, updates, ...
... difference if change in db is needed or not
... we need to support both cases

<Seema> Oracle presentation covers both "db" and "no db" cases

<hhalpin> hmmm...

ahmed: interested to get to know difficult subqueries for mysql

soeren: question seems not to be important
... there are some issues with every db

<ericP> http://lists.w3.org/Archives/Public/public-rdb2rdf-wg/2010Jan/att-0050/RDB2RDFTest.scala

soeren: but there has been spent a lot of time on db development

ericp: interested if those sql queries could make troubles

<hhalpin> i think we can continue for a bit till discussion ends

ahmed: subquery is no issue at any level
... problem is the optimiser for specific types of queries

<mhausenblas> I'd love to see a write-up of today's discussion :)

ahmed: create a list of typical queries generated from sparql to sql to figure out the difficult ones

<mhausenblas> Michael: I like today's discussion - eventually we start to disagree

ahmed: start collecting cases on the wiki

ericp: sql generated from rdb2rdf translations, sql that is a "show stopper"

ahmed: would like to see an approach that does not touch the db

<ericP> GeneratedSQL and PotentialSQLIssues

<mhausenblas> thanks to ericP for kicking-off the Wiki stuff

souri: oracle presentation mentioned that in the case db is not touched

<Ahmed> Bye.

<hhalpin> bye

<mhausenblas> adjourned

- DRAFT -

RDB2RDF Working Group Teleconference

26 Jan 2010

Attendees

Contents

roll call

approve minutes

R2RML Mapping aspects

Summary of Action Items

Scribe.perl diagnostic output