[PATCH] O5int for interactivity

[PATCH] O5int for interactivity

[Con Kolivas]

More interactivity work for audio and X smoothness. I have fixed all my test
 cases and need feedback about others to develop beyond this.

Changes
The idle code now gives just under interactive state based on the runtime
 instead of min_sleep_avg - minor startup speed improvement.

Tasks that drop their priority while running are now put to the end of the
 queue to continue their timeslice. Fixes a little flutter when tasks are
cpu hogs for short periods (eg mozilla).

Tasks that are complete cpu hogs are put on the expired array every time they
 run out of timeslice.

Con

Patch against 2.5.75-mm1 available here:
http://kernel.kolivas.org/2.5

and here:

diff -Naurp linux-2.5.75-mm1/kernel/sched.c linux-2.5.75-test/kernel/sched.c
--- linux-2.5.75-mm1/kernel/sched.c	2003-07-13 00:21:30.000000000 +1000
+++ linux-2.5.75-test/kernel/sched.c	2003-07-14 22:13:51.000000000 +1000
@@ -388,19 +388,19 @@ static inline void __activate_task(task_
 static inline void activate_task(task_t *p, runqueue_t *rq)
 {
 	long sleep_time = jiffies - p->last_run - 1;
+	long runtime = jiffies - p->avg_start;
 
 	if (sleep_time > 0) {
-		unsigned long runtime = jiffies - p->avg_start;
-
 		/*
 		 * Tasks that sleep a long time are categorised as idle and
-		 * will get just under interactive status with a small runtime
-		 * to allow them to become interactive or non-interactive rapidly
+		 * will get just under interactive status
 		 */
 		if (sleep_time > MIN_SLEEP_AVG){
-			p->avg_start = jiffies - MIN_SLEEP_AVG;
-			p->sleep_avg = MIN_SLEEP_AVG * (MAX_BONUS - INTERACTIVE_DELTA - 1) /
-				MAX_BONUS;
+			if (runtime > MAX_SLEEP_AVG){
+				runtime = MAX_SLEEP_AVG;
+				p->avg_start = jiffies - runtime;
+			}
+			p->sleep_avg = runtime * (MAX_BONUS - INTERACTIVE_DELTA - 1) / MAX_BONUS;
 		} else {
 			/*
 			 * This code gives a bonus to interactive tasks.
@@ -1278,11 +1278,7 @@ void scheduler_tick(int user_ticks, int
 	spin_lock(&rq->lock);
 	/*
 	 * The task was running during this tick - update the
-	 * time slice counter and the sleep average. Note: we
-	 * do not update a thread's priority until it either
-	 * goes to sleep or uses up its timeslice. This makes
-	 * it possible for interactive tasks to use up their
-	 * timeslices at their highest priority levels.
+	 * time slice counter and the sleep average.
 	 */
 	if (p->sleep_avg)
 		p->sleep_avg--;
@@ -1309,12 +1305,21 @@ void scheduler_tick(int user_ticks, int 
 		p->time_slice = task_timeslice(p);
 		p->first_time_slice = 0;
 
-		if (!TASK_INTERACTIVE(p) || EXPIRED_STARVING(rq)) {
+		if (!TASK_INTERACTIVE(p) || EXPIRED_STARVING(rq) || !p->sleep_avg) {
 			if (!rq->expired_timestamp)
 				rq->expired_timestamp = jiffies;
 			enqueue_task(p, rq->expired);
 		} else
 			enqueue_task(p, rq->active);
+	} else if (unlikely(p->prio < effective_prio(p))){
+		/*
+		 * Tasks that have lowered their priority are put to the end
+		 * of the active array with their remaining timeslice
+		 */
+		dequeue_task(p, rq->active);
+		set_tsk_need_resched(p);
+		p->prio = effective_prio(p);
+		enqueue_task(p, rq->active);
 	}
 out_unlock:
 	spin_unlock(&rq->lock);

Re: [PATCH] O5int for interactivity

[Felipe Alfaro Solana]

On Mon, 2003-07-14 at 14:32, Con Kolivas wrote:
> More interactivity work for audio and X smoothness. I have fixed all my test
>  cases and need feedback about others to develop beyond this.
> 
> Changes
> The idle code now gives just under interactive state based on the runtime
>  instead of min_sleep_avg - minor startup speed improvement.
> 
> Tasks that drop their priority while running are now put to the end of the
>  queue to continue their timeslice. Fixes a little flutter when tasks are
> cpu hogs for short periods (eg mozilla).
> 
> Tasks that are complete cpu hogs are put on the expired array every time they
>  run out of timeslice.

Hmmm... Starvation is back for me (Pentium III 700Mhz + ACPI):

1. Log on to X/KDE
2. Launch Konqueror
3. Launch XMMS
4. Make XMMS play a song
5. Move the Konqueror window all over the desktop.

Step 5 causes XMMS to completely starve for exactly 5 seconds. After
those 5 seconds, the XMMS priority gets adjusted and sound comes back
from my speakers.

Another way to starve XMMS for 5 seconds:

1. Launch XMMS
2. Make it play
3. Run a standard CPU hogger: "while true; do a=2; done"

Step 3 will make XMMS starve for exactly 5 seconds. Also, during those 5
seconds, X is completely jerky and moving the mouse makes the pointer
goes jumping all over the screen.

Do you want additional information? Any patch trying?
Thanks!

Re: [PATCH] O5int for interactivity

[Con Kolivas]

On Tue, 15 Jul 2003 01:25, Felipe Alfaro Solana wrote:
> On Mon, 2003-07-14 at 14:32, Con Kolivas wrote:
> > More interactivity work for audio and X smoothness. I have fixed all my
> > test cases and need feedback about others to develop beyond this.
> >
> > Changes
> > The idle code now gives just under interactive state based on the runtime
> >  instead of min_sleep_avg - minor startup speed improvement.
> >
> > Tasks that drop their priority while running are now put to the end of
> > the queue to continue their timeslice. Fixes a little flutter when tasks
> > are cpu hogs for short periods (eg mozilla).
> >
> > Tasks that are complete cpu hogs are put on the expired array every time
> > they run out of timeslice.
>
> Hmmm... Starvation is back for me (Pentium III 700Mhz + ACPI):
>
> 1. Log on to X/KDE
> 2. Launch Konqueror
> 3. Launch XMMS
> 4. Make XMMS play a song
> 5. Move the Konqueror window all over the desktop.
>
> Step 5 causes XMMS to completely starve for exactly 5 seconds. After
> those 5 seconds, the XMMS priority gets adjusted and sound comes back
> from my speakers.
>
> Another way to starve XMMS for 5 seconds:
>
> 1. Launch XMMS
> 2. Make it play
> 3. Run a standard CPU hogger: "while true; do a=2; done"
>
> Step 3 will make XMMS starve for exactly 5 seconds. Also, during those 5
> seconds, X is completely jerky and moving the mouse makes the pointer
> goes jumping all over the screen.
>
> Do you want additional information? Any patch trying?
> Thanks!

Great thanks for test. Can you try reverting this:


diff -Naurp linux-2.5.75-mm1/kernel/sched.c linux-2.5.75-test/kernel/sched.c
--- linux-2.5.75-mm1/kernel/sched.c	2003-07-13 00:21:30.000000000 +1000
+++ linux-2.5.75-test/kernel/sched.c	2003-07-14 22:13:51.000000000 +1000
@@ -388,19 +388,19 @@ static inline void __activate_task(task_
 static inline void activate_task(task_t *p, runqueue_t *rq)
 {
 	long sleep_time = jiffies - p->last_run - 1;
+	long runtime = jiffies - p->avg_start;
 
 	if (sleep_time > 0) {
-		unsigned long runtime = jiffies - p->avg_start;
-
 		/*
 		 * Tasks that sleep a long time are categorised as idle and
-		 * will get just under interactive status with a small runtime
-		 * to allow them to become interactive or non-interactive rapidly
+		 * will get just under interactive status
 		 */
 		if (sleep_time > MIN_SLEEP_AVG){
-			p->avg_start = jiffies - MIN_SLEEP_AVG;
-			p->sleep_avg = MIN_SLEEP_AVG * (MAX_BONUS - INTERACTIVE_DELTA - 1) /
-				MAX_BONUS;
+			if (runtime > MAX_SLEEP_AVG){
+				runtime = MAX_SLEEP_AVG;
+				p->avg_start = jiffies - runtime;
+			}
+			p->sleep_avg = runtime * (MAX_BONUS - INTERACTIVE_DELTA - 1) / MAX_BONUS;
 		} else {
 			/*
 			 * This code gives a bonus to interactive tasks.