- From: W.M. Jaworski <wmj@gen-strategies.com>
- Date: Fri, 22 Jun 2001 13:05:15 -0400
- To: "www-rdf-dspace" <www-rdf-dspace@w3.org>, <dspace-code@MIT.EDU>, "Peter Breton" <pbreton@MIT.EDU>
[Peter Breton] The beauty of the triple store mechanism is that you can accomodate ALL the data in it. [wmj] Other store mechanisms (for example relational) also "accomodate ALL the data in it." [Peter Breton] the roundtrips from client/middleware to RDBMS are likely to be some of the most expensive operations [wmj] If the 'client/middleware' is not RDF, than is it another notation? Is RDF notation a technology looking for the applications? [Peter Breton] 3) Another road: since queries on an unbounded triple store .... [wmj] It seems that the problem is solved by "Associative Model of Data™ - http://www.lazysoftware.com BTW I am not an opponent of RDF, I am an outsider looking for insights and knowledge. Respectfully, WMJ -----Original Message----- From: www-rdf-dspace-request@w3.org [mailto:www-rdf-dspace-request@w3.org]On Behalf Of Peter Breton Sent: Friday, June 22, 2001 10:24 AM To: dspace-code@MIT.EDU; www-rdf-dspace Subject: some more RDF thoughts 1) On storage: The beauty of the triple store mechanism is that you can accomodate ALL the data in it. You don't need separate mechanisms to store schema information, taxonomies, and the like: it's all triples, all the way down. (apologies to those who find this excruciatingly obvious!). 2) On scalability: It's difficult for me to see how SQL queries on an unbounded generic triple store will _ever_ scale. Scalability in an RDBMS is generally achieved precisely by non-generic methods: pulling data into well-known columns which can be indexed. There are some tricks that might help, however: * Perhaps use graph-oriented indexes on databases which support them? * Somehow offload the RDF processing to the database, since the roundtrips from client/middleware to RDBMS are likely to be some of the most expensive operations I think Postgres could be hacked to do one or both of these. 3) Another road: since queries on an unbounded triple store may always be problematic, separate the triple stores. One possibility running through my head is a "double triple store". The first triple store serves as a cache for the second one (and there could also be an in-memory cache, like the Collego folks have). The first triple store could include (as triples, natch!) a description of all the info it has. "Efficiency" (really, efficient queries) could be thus be achieved by simply migrating or copying data to one or more triple store caches. And since the mapping data is tiny, the overhead of figuring out which cache to query should be minimal (and could be done in-memory). I say "cache" above, but it could also be migration. Peter
Received on Friday, 22 June 2001 12:53:27 UTC