> David Feuer <feuer@his.com> writes:
>
> > Why is this better than nice-19?
>
> It totally removes the job from the OS's run queue, so it takes _no_ CPU
> time when the load crosses a threshold. When the load drops below a second
> threshold, it puts it back on. So, if your loadwatched job holds the load
> at 1.0 normally, you set the highpoint for something like 1.15. When
> anything significant starts, the load goes over 1.15, and the loadwatched
> job totally stops (with SIGSTOP). Whatever loaded the system will
> presumably now keep the load over the second point, say 0.15, and it will
> not have the loadwatched job stealing any cycles from it. Once the load
> drops below 0.15, the loadwatched job is sent a SIGCONT.
>
> As mentioned elsewhere in this thread, a nice 19 job still gets 5% of the
> CPU. And, if that job is using resources like lots of memory or doing I/O,
> it can have a disproportional impact on the system. Even for a CPU bound
> job like rc5des, some people can notice it running at nice 19 because of
> slightly slower interactive performance. If you're one of them, loadwatch
> works pretty nicely for getting rid of that small lag.
>
> Later,
> Dale
> --
> +-------------------- finger for pgp public key ---------------------+
> | Dale E. Martin | Clifton Labs, Inc. | Senior Computer Engineer |
> | dmartin@clifton-labs.com | http://www.clifton-labs.com |
> +----------------------------------------------------------------------+
What do you mean by 5% of the CPU? It sounds to me like Linux could use a
nice 20...... still some tiny overhead for scheduling, but I'm guessing not
more than sysload polling.....
--______________________________ / David Feuer \ | dfeuer@binx.mbhs.edu | | feuer@his.com | \ david@feuer.his.com / -----------------------------
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/