[Barrelfish-users] Is Barrelfish tickless?
Timothy Roscoe
troscoe at inf.ethz.ch
Thu May 31 10:20:00 CEST 2012
To some extent, the use of a periodic timer is an engineering
convenience. We have been known to change the default scheduler (RBED)
to a much simpler, round-robin scheduler for debugging, and RR simply
uses a periodic timer.
RBED, our preferred scheduler, actually has no real notion of periodic
interrupts (and rightly so). Consequently, our use of periodic ticker
for RBED is simply an approximation of a programmable interrupt timer.
Long term, we'll move to a real timer.
-- Mothy
On 05/31/2012 09:18 AM, Kornilios Kourtis wrote:
> Hi Hakan,
>
> On Wed, May 30, 2012 at 02:43:57PM -0600, Hakan Akkan wrote:
>> Just wondering if the Barrelfish kernel is tickless or not. Was this a
>> design criteria? By tickless I mean if the kernel receives and
>> services periodic timer interrupts on the CPUs where an application is
>> supposed to run.
>
> The barrelfish kernel (i.e., the CPU driver) is not tickless by default.
> It will use periodic timer interrupts and, based on the scheduling
> policy, decide whether to preempt the application or not.
>
> There is a config option called oneshot_timer (which is disabled by
> default) where the timer is programmed based on the time of the next
> event -- e.g., the time when the scheduling timeslice of the running
> application will expire. Note, however, that this option has not been
> heavily tested.
>
> cheers,
> Kornilios.
>
More information about the Barrelfish-users
mailing list