> looks fine for me.
>
> Just one extra idea... May be we can add some flag into
> SelectorHandler: "isExecPendingIoAsync" or similar to be able to
> chose if we want to execute pending cancel operations in the same or
> dedicated thread. So we will be able to switch the mode quickly, in
> case we will find any perf. regression.
>
> I will add a flag for that.
> The only concern imo is that the events happen sometimes later, if
> pool is already heavily loaded it might be too late.
I don't mean, that want to change this value at runtime, but on
initialization phase.
Alexey.
>
>
>
> Great job!!!
>
> Thank you.
>
> WBR,
> Alexey.
>
>
> On Apr 3, 2009, at 2:08 , gustav trede wrote:
>
> Hello,
>
> Please feedback on patch for moving socket close and attachment
> triggered IO from selector thread to workers at postSelect.
> The SelectionKey.cancel() is of course still done in the selecting
> thread.
> Currently no Runnable wrappers are needed per task due to the
> affected attachment objects was made Runnable.
>
> --
> regards
> gustav trede
>
>
> <
> patch
> .diff
> >---------------------------------------------------------------------
> To unsubscribe, e-mail: dev-unsubscribe_at_grizzly.dev.java.net
> For additional commands, e-mail: dev-help_at_grizzly.dev.java.net
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: dev-unsubscribe_at_grizzly.dev.java.net
> For additional commands, e-mail: dev-help_at_grizzly.dev.java.net
>
>
>
>
> --
> regards
> gustav trede
>
>