Triplestore write performance and locking

I have seen lots of triplestore benchmarks which focus on read and query performance. However, the application which I am looking at moving to a triplestore has very dynamic data and I am very concerned about write performance, and locking.

For example, I have heard that Jena/TDB is only usable with a write-once, read-many locking model in force. This permits several reader SPARQL queries can run at the same time. However, when a write operation takes place to create/update triples, the write operation cannot be serviced until any existing running SPARQL queries have completed. Furthermore, no read operations can take place (on any triples in the model开发者_StackOverflow社区) whilst the write operation is in progress. You can imagine a bulk data copy operation blocking any SPARQL queries for a long time.

Well, you can imagine this locking model would have severe consequences for my application!

Has anyone done any comparisons of the different triple stores out there at this level?

Even if you could comment on the triple store that you are using yourself, that would still be extremely useful for me.

For high write performance, you might also (depending on your application) want to consider a NoSQL datastore such as Apache Cassandra.