[coyotos-dev] Sleep, wakeup, and persistence

[Valerio Bellizzomi]

On 15/09/2007, at 2.42, Marcus Brinkmann wrote:

>At Fri, 14 Sep 2007 22:53:15 +0200,
>"Valerio Bellizzomi" <devbox at selnet.org> wrote:
>> The mainboard clock counts faster than real time, so when NTP time is
>> updated, the real time actually goes backwards.
>
>The typical way to implement synchronization back in time while
>keeping the clock monotonous is to slow down the system clock until it
>is correct again.
>
>Naturally this works best if you do this not too infrequently, because
>the time it requires to synchronize the clock is at least the existing
>clock drift (and typically more than that, because you can not afford to
>keep the system clock at a complete stand still for an extended period
>of time).

With the clock at 50% it should require (drift * 2) plus some small time
to manipulate the clock.
Probably it is affordable to slow down the clock permanently to reduce the
drift. There might also be a clock value that keeps the wall time
approximately accurate, in that case one can synchronize less frequently.
Learning the right value should not be that difficult with NTP.
But I don't know if it is useful enought to implement.

>
>Thanks,
>Marcus
>
>_______________________________________________
>coyotos-dev mailing list
>coyotos-dev at smtp.coyotos.org
>http://www.coyotos.org/mailman/listinfo/coyotos-dev