For the EJB 3.1 TimerService there are two goals:
1. EJB 3.1 functionality
2. true clustering
For the EJB 3.1 functionality refer to chapter 18 of the spec.
Right now the timer service doesn't do real clustering. It's setup as a HA-singleton, which makes for an imbalanced cluster.
The new timer service should also make use of a Hibernate backend. This to eliminate any problems with different databases and SQL schemas.
It would be nice to have timers back on stateful objects. Since EJB 2.1 entities are eliminated, stateful beans become the obvious choice. (Note that this is in violation of the spec.)
So in the end I would like to see two components:
1. a 'light' version that implements the spec
2. a 'heavy' version that extends it with clustering features
Both components should be runnable (and testable) in a standalone environment.