> As far as I understand the billing system, it works that way:
> If a billable process calls a service from a non-billable
> process, the ticks the non-billable process consumes is billed
> on the billable process in system-time. Correct?

Yes. Unfortunately, accounting in a multiserver system is complicated, and
this is not bullet-proof. Suppose process A requested I/O and the driver
is blocked waiting for an interrupt, process B may be scheduled in
between. Now, when the interrupt occurs, the bill pointer points to user
process B, while the driver continues doing work on behalf of process A.

> 1) A common case for this example would be
> a user process calling a server, right?

Yes.

> 2) Why are there 2 different billing systems,
> is this used in the scheduling or just for
> statistics?

There is billing for scheduling: the quantum of a user process is
decremented; when it becomes zero, another process is scheduled in a
round-robin fashion.

Furthermore, user and system times are accumulated for each process. (This
is done in the kernel. The PM also accumulates the total child user and
system times at the parent whenever a process exits.) These times are, for
example, used for the 'time' utility.

Jorrit