[RFC] renicing X

[RFC] renicing X

[Nick Piggin]

My scheduler patch really benefits a lot from renicing X. I
think its because it nices more nicely. Any reasons why this
might be a bad idea?

Re: [RFC] renicing X

[Felipe Alfaro Solana]

On Mon, 2003-08-25 at 18:00, Nick Piggin wrote:
> My scheduler patch really benefits a lot from renicing X. I
> think its because it nices more nicely. Any reasons why this
> might be a bad idea?

Well, not for me... Although renicing X with Con patches makes X feel
horrible, with your patches is not as horrible. However, I feel X much
smoother with X reniced at +0. Renicing X at -20, for example, may
reduce mouse cursor jumpiness under load, but makes X feel a little
jerky in general (window movement is not as smooth as with X niced at
+0). This is, however, based on subjective testing, not actual numbers.
But for interactivity, most of the time it's the subjective feeling of
the user about the system what matters, not numbers.

And now we're talking about sched-policy-7a: under heavy load, spawning
new processes still takes twice the time it takes when the system is
under no load. For example, spawning a new Konsole session (not a new
Konsole process, but a new session tab inside Konsole) takes approx. 1
second on my P3-700Mhz. However, with sched-policy-7a and under heavy
load (the mad while true; do a=2; done loop), it takes more than 2
seconds.

In general, sched-policy-7a feels extremely smooth and responsive in
general but, for me, Con patches offer the smoothest X experience I have
ever felt until date. Anyways, I will keep testing your patches and I
greatly encourage you to keep improving them. It's always good to have
diversity :-)

Re: [RFC] renicing X

[Nick Piggin]

Felipe Alfaro Solana wrote:

>On Mon, 2003-08-25 at 18:00, Nick Piggin wrote:
>
>>My scheduler patch really benefits a lot from renicing X. I
>>think its because it nices more nicely. Any reasons why this
>>might be a bad idea?
>>

Hi Felipe,
Sorry I can't mail you directly. Some spam filter doesn't like me.

>
>Well, not for me... Although renicing X with Con patches makes X feel
>horrible, with your patches is not as horrible. However, I feel X much
>smoother with X reniced at +0. Renicing X at -20, for example, may
>reduce mouse cursor jumpiness under load, but makes X feel a little
>jerky in general (window movement is not as smooth as with X niced at
>+0). This is, however, based on subjective testing, not actual numbers.
>

Hmm interesting. Might be a bug...

>
>But for interactivity, most of the time it's the subjective feeling of
>the user about the system what matters, not numbers.
>

Yep

>
>And now we're talking about sched-policy-7a: under heavy load, spawning
>new processes still takes twice the time it takes when the system is
>under no load. For example, spawning a new Konsole session (not a new
>Konsole process, but a new session tab inside Konsole) takes approx. 1
>second on my P3-700Mhz. However, with sched-policy-7a and under heavy
>load (the mad while true; do a=2; done loop), it takes more than 2
>seconds.
>

OK thanks. I'll try to work on this.

>
>In general, sched-policy-7a feels extremely smooth and responsive in
>general but, for me, Con patches offer the smoothest X experience I have
>ever felt until date. Anyways, I will keep testing your patches and I
>greatly encourage you to keep improving them. It's always good to have
>diversity :-)
>

Well thats good. Thanks very much.