>> Even better (if applicable) would be to have fast thread-local counters,
>> with a slow background phase which occasionally tallies them into a trailing
>> global counter.

> That reminds me of an old question of mine.  In the case where a
> thread has been created from an object whose class is a subclass of
> Thread, is Thread.currentThread() guaranteed to give you back the same
> object?  IOW, if I say:

> class MyThread extends Thread { ... }
> myThread = new MyThread();
> myThread.start();
> ...

> then in the new thread, is Thread.currentThread() always == to the
> original value of myThread?
yes
>  If so, that provides a different approach
> to thread-local state, where the state is held in the instance
> variables of MyThread objects,