Do you know the talents and weak points of the triplestore database, for example Mulgara? By which situations would its performance be considerably not the same as what traditional relational database?
I am thinking particularly about situations in which you have data within the form
namespace:predicate=value. Triplestores would appear to help keep the information in the "natural" form, but I must determine if you will find specific advantages beyond that.
I am the creator of Redland including native triple stores in C plus I have used a few of the Java systems like Jena and Sesame.
It's difficult to express more not understanding what your condition is however in general native triple stores are more effective in case your problem space/query pattern is much more 'graphy', instead of 'objecty'. Objecty means you often always manipulate - set/obtain the same pattern of information every time - for the reason that situation you may as well use ORM or any other object-relational mappping along with a fixed schema. In case your problem and schema is much more dynamic - 'Graphy' - you might enjoy the triple store approach where it's simpler to hop over the data than typically inside a RDBMS world (also known as joins).
Obviously there's even the document-approach too for example CouchDB but that is difficult to know in the info you've given.
I suppose you are asking about persistence of triples (and perhaps graphs via quads) in which the storage will be an "native" triple store (for instance, TDB) or perhaps a relational database having a triple store overlay (like SDB).
Should you a more sophisticated in your needs, I am sure Stack Overflow people can provide a far more directed answer, but simply off the top my mind:
- Native triple stores routinely have better performance when how big the dataset will get large. See some numbers from the W3C.
- Native triple stores can provide in-store reasoning/inferencing (like Oracle's semantic toolkit, even though it does not fully support SPARQL).
- Triples within an RDBMS could be "easily" built-into traditional confirming tools (like Jasper Reviews) since query all of them with SQL.