[Click] NSClick - performance upgrade for NS-2
Björn Lichtblau
lichtbla at informatik.hu-berlin.de
Wed Jan 25 15:52:33 EST 2012
Hi,
On 25.01.2012 21:15, Eddie Kohler wrote:
>
> It might be reasonable to set SLEEPINESS_TRIGGER to 0 in ns. I don't
> love this approach, though.
>
> The sleepiness trigger is addressing the problem of deceptive idleness
> in scheduling. When in heavy usage, it was faster to keep an element
> scheduled than to unschedule it, after which the Click thread might go
> to sleep as a whole.
>
> Click's inner scheduling loop is quite fast. My guess is that NS's is
> not as fast. If the problem was event pileup, then we just should fix
> that problem.
>
> Eddie
thanks, i will try with SLEEPINESS_TRIGGER, sounds like that is all.
Click can't run too fast, Click and simulator are synchronized, running
serialized. Time in Click won't advance until routerthread returns to
the simulator, and the simulator calls Click exactly at that time it
requested with SIMCLICK_SCHEDULE.
Regarding the commit i get strange results, but i have to recheck. Does
this patch try to also solve the issue Lalith raised: "Patch for
nanosecond scheduling precision in ns-3-click" ?
Regards, Björn
More information about the click
mailing list