Re: Getting Started with Hurd-L4

[Espen Skoglund]

[Sam Mason]
> Marcus Brinkmann wrote:
>> I don't remember something like "scheduler activations reference",
>> but certainly we are expected to (and actually need to) implement a
>> user-space scheduler.

> "Scheduler activations" are for the scheduling of the users threads
> inside of a task - which, in general, have nothing to do with the
> threads as seen by the kernel.  I assume what you mean is the
> scheduler for the kernel-level threads.  I'd personally expect this
> to be implemented quite late on, with a very basic scheduling
> algorithm being used initially.

One of the main reasons for introducing scheduler activations was the
assumption that kernel-level threads are pretty heavy weight, i.e.,
they consume loads of resources and context switching between them are
expensive.

In L4 land we tend to view kernel-level threads as pretty light weight
(or at least not particularly heavy weight).  They do not consume that
many resources (although the TCBs can on some architectures be pretty
big), and context swicthing overhead can be overcome by using lazy
context switching [1].  Just because scheduler activations are
suitable for monolithic systems like FreeBSD or NetBSD (these systems
chosen because they actually implement it) does not mean that they are
suitable for L4.

        eSk


[1] J. Liedtke and H. Wenske.  Lazy Process Switching.  8th Workshop
on Hot Topics in Operating Systems (HotOS-VIII), Elmau/Oberbayern,
Germany, May 2001