Hi,

a few days ago I wrote a little test, that registers some installations in a concurrent fashion against a running UPS. Here is a polished version: https://gist.github.com/matzew/bb2c11bf55e2e5c7334d

The RESTful endpoints are behaving fine (they return 200) and deliver internally the tasks to a new thread, using our Async EJB. yay!

Of course, on different databases I noticed different results:

* POSTGRES
With Postgres I got reasonalbe/good results:
16 threads/1000 loops -> unit test took 50 seconds. (16k devices)
32 threads/1000 loops -> unit test took 120 seconds. (32k devices)

==> would be nice if we could improve this (e.g. with better JPQL or something like that)


* H2 (WARNING - not a production database)
Writing to the H2 DB is (extremely) slow, regardless if in-memory or file-system :)
So, when the Unit test was done/finised, the writing to the (H2) database still goes on, for quite some time.
Using the H2 filesystem storage, I noticed some of these log messages -> https://gist.github.com/matzew/41635a254cbec9057e77

Sure, it's H2 and we only use that during our development and test execution but I want at least mention the above behavior

* MYSQL
TBD

Oh, another thing I thought about, perhaps it would be possible to use (embedded) Postgress/MySQL during our tests with these guys:
* https://github.com/NessComputing/components-ness-pg
* mysql-connector-mxj


Any thoughts ? 
-Matthias

--
Matthias Wessendorf

blog: http://matthiaswessendorf.wordpress.com/
sessions: http://www.slideshare.net/mwessendorf
twitter: http://twitter.com/mwessendorf