On Mon, Jun 13, 2016 at 1:40 AM, Thomas Gleixner <[email protected]> wrote: > The current timer wheel has some drawbacks:
... > This series has also preparatory patches for changing the NOHZ timer handling > from the current push to a pull model. Currently we decide at timer enqueue > time on which cpu we queue the timer. This is exceptionally silly because > there is no way to predict at enqueue time which cpu will be idle when the > timer expires. Given the fact that most timers are canceled or rearmed before > expiry this is even more silly. We trade a expensive decision and cross cpu > access for a very doubtful benefit. > Also many TCP timers are armed from softirq handler interrupting idle task, so scheduler believes current cpu is idle . This patch series is really nice Thomas.
