Reproducible terrible interactivity since 2.5.64bk2
(Please CC me your answers, thank you)
I'm seeing serious interactivity problems in 2.5.65, 2.5.66, 2.5.65-mm4 and 2.5.64-bk2. I couldn't reproduce it on 2.5.64, 2.5.64-bk1.
I noticed that when I was archiving a directory with some CD ISO images with tar & bzip2. After a while (~5 minutes) many programs took a long time to respond, XMMS paused for several seconds, starting simple programs (mc, ps, ls) took sometimes even 40s. Disk is idle most of the time.
To reproduce I wrote a simple script:
#!/bin/sh
cat cedo.iso | bzip2 > /tmp/cedo.iso.bz2 &
for i in seq 1 20; do
sleep 30
date
ps x
date
echo -----
done
... where cedo.iso is a 700MB CD ISO image.
When run on 2.5.64-bk2 (and no X running) I've got this output:
Wed Mar 26 10:47:50 CET 2003
PID TTY STAT TIME COMMAND 325 tty2 S 0:00 -bash 713 tty2 R 0:00 /bin/sh ./test-interactivity.sh 714 tty2 R 0:00 tee inter-2.5.64bk2.log 715 tty2 S 0:00 cat cedo.iso 716 tty2 R 0:33 bzip2 720 tty2 R 0:00 ps x Wed Mar 26 10:47:50 CET 2003 ----- Wed Mar 26 10:48:22 CET 2003 PID TTY STAT TIME COMMAND 325 tty2 S 0:00 -bash 713 tty2 S 0:00 /bin/sh ./test-interactivity.sh 714 tty2 S 0:00 tee inter-2.5.64bk2.log 715 tty2 S 0:00 cat cedo.iso 716 tty2 R 1:05 bzip2 724 tty2 R 0:00 ps x Wed Mar 26 10:48:23 CET 2003 ----- Wed Mar 26 10:48:55 CET 2003 PID TTY STAT TIME COMMAND 325 tty2 S 0:00 -bash 713 tty2 S 0:00 /bin/sh ./test-interactivity.sh 714 tty2 S 0:00 tee inter-2.5.64bk2.log 715 tty2 S 0:00 cat cedo.iso 716 tty2 R 1:38 bzip2 728 tty2 R 0:00 ps x Wed Mar 26 10:48:55 CET 2003 ----- Wed Mar 26 10:49:28 CET 2003 PID TTY STAT TIME COMMAND 325 tty2 S 0:00 -bash 713 tty2 R 0:00 /bin/sh ./test-interactivity.sh 714 tty2 S 0:00 tee inter-2.5.64bk2.log 715 tty2 S 0:00 cat cedo.iso 716 tty2 R 2:10 bzip2 732 tty2 R 0:00 ps x Wed Mar 26 10:49:28 CET 2003 ----- Wed Mar 26 10:50:01 CET 2003 PID TTY STAT TIME COMMAND 325 tty2 S 0:00 -bash 713 tty2 S 0:00 /bin/sh ./test-interactivity.sh 714 tty2 S 0:00 tee inter-2.5.64bk2.log 715 tty2 S 0:00 cat cedo.iso 716 tty2 R 2:42 bzip2 736 tty2 R 0:00 ps x Wed Mar 26 10:50:01 CET 2003 ----- Wed Mar 26 10:50:34 CET 2003 PID TTY STAT TIME COMMAND 325 tty2 S 0:00 -bash 713 tty2 S 0:00 /bin/sh ./test-interactivity.sh 714 tty2 S 0:00 tee inter-2.5.64bk2.log 715 tty2 S 0:00 cat cedo.iso 716 tty2 R 3:15 bzip2 740 tty2 R 0:00 ps x Wed Mar 26 10:50:34 CET 2003 ----- Wed Mar 26 10:51:06 CET 2003 PID TTY STAT TIME COMMAND 325 tty2 S 0:00 -bash 713 tty2 R 0:00 /bin/sh ./test-interactivity.sh 714 tty2 S 0:00 tee inter-2.5.64bk2.log 715 tty2 S 0:00 cat cedo.iso 716 tty2 R 3:47 bzip2 744 tty2 R 0:00 ps x Wed Mar 26 10:51:06 CET 2003 ----- Wed Mar 26 10:51:39 CET 2003 PID TTY STAT TIME COMMAND 325 tty2 S 0:00 -bash 713 tty2 S 0:00 /bin/sh ./test-interactivity.sh 714 tty2 S 0:00 tee inter-2.5.64bk2.log 715 tty2 S 0:00 cat cedo.iso 716 tty2 R 4:19 bzip2 748 tty2 R 0:00 ps x Wed Mar 26 10:51:39 CET 2003 ----- Wed Mar 26 10:52:12 CET 2003 PID TTY STAT TIME COMMAND 325 tty2 S 0:00 -bash 713 tty2 S 0:00 /bin/sh ./test-interactivity.sh 714 tty2 S 0:00 tee inter-2.5.64bk2.log 715 tty2 S 0:01 cat cedo.iso 716 tty2 R 4:52 bzip2 752 tty2 R 0:00 ps x Wed Mar 26 10:52:12 CET 2003 ----- Wed Mar 26 10:52:45 CET 2003 PID TTY STAT TIME COMMAND 325 tty2 S 0:00 -bash 713 tty2 R 0:00 /bin/sh ./test-interactivity.sh 714 tty2 S 0:00 tee inter-2.5.64bk2.log 715 tty2 S 0:01 cat cedo.iso 716 tty2 R 5:24 bzip2 756 tty2 R 0:00 ps x Wed Mar 26 10:52:45 CET 2003 ----- Wed Mar 26 10:53:17 CET 2003 PID TTY STAT TIME COMMAND 325 tty2 S 0:00 -bash 713 tty2 S 0:00 /bin/sh ./test-interactivity.sh 714 tty2 S 0:00 tee inter-2.5.64bk2.log 715 tty2 S 0:01 cat cedo.iso 716 tty2 R 5:57 bzip2 760 tty2 R 0:00 ps x Wed Mar 26 10:53:17 CET 2003 ----- Wed Mar 26 10:53:50 CET 2003 PID TTY STAT TIME COMMAND 325 tty2 S 0:00 -bash 713 tty2 S 0:00 /bin/sh ./test-interactivity.sh 714 tty2 S 0:00 tee inter-2.5.64bk2.log 715 tty2 S 0:01 cat cedo.iso 716 tty2 R 6:29 bzip2 764 tty2 R 0:00 ps x Wed Mar 26 10:53:50 CET 2003 ----- Wed Mar 26 10:54:23 CET 2003 PID TTY STAT TIME COMMAND 325 tty2 S 0:00 -bash 713 tty2 R 0:00 /bin/sh ./test-interactivity.sh 714 tty2 S 0:00 tee inter-2.5.64bk2.log 715 tty2 S 0:01 cat cedo.iso 716 tty2 R 7:01 bzip2 768 tty2 R 0:00 ps x Wed Mar 26 10:54:23 CET 2003 ----- Wed Mar 26 10:54:56 CET 2003 PID TTY STAT TIME COMMAND 325 tty2 S 0:00 -bash 713 tty2 S 0:00 /bin/sh ./test-interactivity.sh 714 tty2 S 0:00 tee inter-2.5.64bk2.log 715 tty2 D 0:01 cat cedo.iso 716 tty2 S 7:41 bzip2 772 tty2 R 0:00 ps x Wed Mar 26 10:55:03 CET 2003 ----- Wed Mar 26 10:55:45 CET 2003 PID TTY STAT TIME COMMAND 325 tty2 S 0:00 -bash 713 tty2 S 0:00 /bin/sh ./test-interactivity.sh 714 tty2 S 0:00 tee inter-2.5.64bk2.log 715 tty2 D 0:01 cat cedo.iso 716 tty2 S 8:50 bzip2 776 tty2 R 0:00 ps x Wed Mar 26 10:56:13 CET 2003 ----- Wed Mar 26 10:56:50 CET 2003 PID TTY STAT TIME COMMAND 325 tty2 S 0:00 -bash 713 tty2 R 0:00 /bin/sh ./test-interactivity.sh 714 tty2 S 0:00 tee inter-2.5.64bk2.log 715 tty2 S 0:01 cat cedo.iso 716 tty2 R 9:27 bzip2 780 tty2 R 0:00 ps x Wed Mar 26 10:56:50 CET 2003 ----- Wed Mar 26 10:57:23 CET 2003 PID TTY STAT TIME COMMAND 325 tty2 S 0:00 -bash 713 tty2 S 0:00 /bin/sh ./test-interactivity.sh 714 tty2 S 0:00 tee inter-2.5.64bk2.log 715 tty2 D 0:02 cat cedo.iso 716 tty2 S 10:23 bzip2 784 tty2 R 0:00 ps x Wed Mar 26 10:57:47 CET 2003 ----- Wed Mar 26 10:58:29 CET 2003 PID TTY STAT TIME COMMAND 325 tty2 S 0:00 -bash 713 tty2 S 0:00 /bin/sh ./test-interactivity.sh 714 tty2 S 0:00 tee inter-2.5.64bk2.log 715 tty2 D 0:02 cat cedo.iso 716 tty2 S 11:09 bzip2 788 tty2 R 0:00 ps x Wed Mar 26 10:58:33 CET 2003 ----- Wed Mar 26 10:59:19 CET 2003 PID TTY STAT TIME COMMAND 325 tty2 S 0:00 -bash 713 tty2 S 0:00 /bin/sh ./test-interactivity.sh 714 tty2 S 0:00 tee inter-2.5.64bk2.log 715 tty2 D 0:02 cat cedo.iso 716 tty2 S 11:59 bzip2 792 tty2 R 0:00 ps x
Wed Mar 26 10:59:24 CET 2003
-----
(I interrupted the script here)
You can see that after less then 8 minutes running date, ps, and date again takes a very long time. When this happens, ps shows cat process in D-state and bzip2 in S-state.
I use Debian Woody on UP Athlon 800MHz, Asus A7V, 384MB RAM, disk WDC WD800JB-00CRA1 connected to on-board Promise 20265, nVidia Geforce2 MX, Realtek RTL-8139C, SB Live. GCC is 2.95.4.
Let me know if you want me to provide more information/testing.
Michal Schmidt.
Quoted 3 lines · Michal Schmidt
I'm seeing serious interactivity problems in 2.5.65, 2.5.66, 2.5.65-mm4 and 2.5.64-bk2.
I couldn't reproduce it on 2.5.64, 2.5.64-bk1.
I'm seeing similar on 2.5.66; xmms pauses when doing disk intensive tasks.
Quoted 4 lines
I use Debian Woody on UP Athlon 800MHz, Asus A7V, 384MB RAM, disk WDC WD800JB-00CRA1 connected to on-board Promise 20265, nVidia Geforce2 MX, Realtek RTL-8139C, SB Live.
GCC is 2.95.4.
My system details: Gentoo Linux 1.2, PIII 1Ghz, 256MB RAM, Ati Rage128, tulip nic, es1371 based sound card, gcc 2.95.3.
I'd be prepared to test any proposed fixes.
thanks,
Andy
The contents of this e-mail and any attachments are confidential and may be legally privileged. If you have received this e-mail and you are not a named addressee, please inform us as soon as possible on +44 118 901 2999 and then delete the e-mail from your system. If you are not a named addressee you must not copy, use, disclose, distribute, print or rely on this e-mail. Any views expressed in this e-mail or any attachments may not necessarily reflect those of Tao's management. Although we routinely screen for viruses, addressees should scan this e-mail and any attachments for viruses. Tao makes no representation or warranty as to the absence of viruses in this e-mail or any attachments. Please note that for the protection of our business, we may monitor and read e-mails sent to and from our server(s).
Quoted 5 lines · Andrew Ebling
I'm seeing similar on 2.5.66; xmms pauses when doing disk intensive tasks.
This may be a different problem. My test is not very disk intensive. It is more CPU intensive (bzip2 compression). The disk is only slightly used when running my testing script.
Michal Schmidt.
Quoted 7 lines · Michal Schmidt
Quoted 2 lines · Andrew Ebling
I'm seeing similar on 2.5.66; xmms pauses when doing disk intensive tasks.
This may be a different problem. My test is not very disk intensive. It is more CPU intensive (bzip2 compression). The disk is only slightly used when running my testing script.
I just experienced the problem very badly when tarring/bzipping hundreds of MB of source files - almost certainly CPU bound on that occasion ;-).
I could do this on 2.4.2x with no problem.
btw. xmms is playing via ALSA, but other apps are also being starved of CPU time - 30 second screen redraws :-/.
Andy
The contents of this e-mail and any attachments are confidential and may be legally privileged. If you have received this e-mail and you are not a named addressee, please inform us as soon as possible on +44 118 901 2999 and then delete the e-mail from your system. If you are not a named addressee you must not copy, use, disclose, distribute, print or rely on this e-mail. Any views expressed in this e-mail or any attachments may not necessarily reflect those of Tao's management. Although we routinely screen for viruses, addressees should scan this e-mail and any attachments for viruses. Tao makes no representation or warranty as to the absence of viruses in this e-mail or any attachments. Please note that for the protection of our business, we may monitor and read e-mails sent to and from our server(s).
Quoted 6 lines · Michal Schmidt
Quoted 3 lines · Andrew Ebling
I'm seeing similar on 2.5.66; xmms pauses when doing disk intensive tasks.
This may be a different problem. My test is not very disk intensive. It is more CPU intensive (bzip2 compression). The disk is only slightly used when running my testing script.
Greetings potential victims :)
Care to see if the attached cures your woes?
This is a mixture of Ingo's last posted plus the scheduler tuning knobs patch (/proc/sys/sched/*). I added three new knobs to watch the effect on different loads. max_accel_slices limits the amount of sleep_time you may add in one activation. retard_prct_slices is a percentage of a slice to deduct from sleep_time each activation (negative feedback for heavy context switchers.. dang irman process_load). force_switch is there because I'm playing :) I didn't do much to the scheduler itself, only made it switch arrays in something closer to a square wave. With the settings as in the patch, and running a kernel build, top and irman, irman reports worst case response times of 150ms for NULL load, 316ms for memory_load, 414 for io_load, and 504ms for process_load.
Anyway, it's attached if you want to play with it ;-)
-Mike
Oh, it's against virgin 2.5.66.
kernel/sched.c +94 -22
@@ -63,17 +63,40 @@* Minimum timeslice is 10 msecs, default timeslice is 100 msecs, * maximum timeslice is 200 msecs. Timeslices get refilled after * they expire. +* +* They are configurable via /proc/sys/sched */ -#define MIN_TIMESLICE ( 10 * HZ / 1000) -#define MAX_TIMESLICE (200 * HZ / 1000) -#define CHILD_PENALTY 50 -#define PARENT_PENALTY 100 -#define EXIT_WEIGHT 3 -#define PRIO_BONUS_RATIO 25 -#define INTERACTIVE_DELTA 2 -#define MAX_SLEEP_AVG (10*HZ) -#define STARVATION_LIMIT (10*HZ) -#define NODE_THRESHOLD 125 + +int min_timeslice = (5 *HZ ) /1000 ; +int max_timeslice = (200 *HZ ) /1000 ; +int child_penalty =50 ; +int parent_penalty =100 ; +int exit_weight =3 ; +int prio_bonus_ratio =25 ; +int interactive_delta =2 ; +int max_sleep_avg =10 *HZ ; +int starvation_limit =1 *HZ ; +int node_threshold =125 ; +int max_accel_slices =5 ;/* Max bonus per activation in slices. */ +int retard_prct_slice =10 ;/* Percent of a slice to deduct. */ +int force_switch =1 ; + +#define MIN_TIMESLICE (min_timeslice) +#define MAX_TIMESLICE (max_timeslice) +#define CHILD_PENALTY (child_penalty) +#define PARENT_PENALTY (parent_penalty) +#define EXIT_WEIGHT (exit_weight) +#define PRIO_BONUS_RATIO (prio_bonus_ratio) +#define INTERACTIVE_DELTA (interactive_delta) +#define MAX_SLEEP_AVG (max_sleep_avg) +#define STARVATION_LIMIT (starvation_limit) +#define NODE_THRESHOLD (node_threshold) +#define TIMESLICE_GRANULARITY (HZ/20 ?: 1) +#define MAX_ACCELERATION (slice) \ + (max_accel_slices ? max_accel_slices * (slice) : max_sleep_avg) +#define NEGATIVE_FEEDBACK (slice) \ + (retard_prct_slice ? retard_prct_slice * (slice) / 100 : 0) +#define FORCE_SWITCH (force_switch)/* * If a task is 'interactive' then we reinsert it in the active
@@ -124,12 +147,17 @@* task_timeslice() is the interface that is used by the scheduler. */ -#define BASE_TIMESLICE (p) (MIN_TIMESLICE + \ - ((MAX_TIMESLICE - MIN_TIMESLICE) * (MAX_PRIO-1-(p)->static_prio)/(MAX_USER_PRIO - 1))) +#define BASE_TIMESLICE (p) \ + (MAX_TIMESLICE * (MAX_PRIO-(p)->static_prio)/MAX_USER_PRIO)static inline unsigned int task_timeslice (task_t *p) { -return BASE_TIMESLICE (p); +unsigned int time_slice =BASE_TIMESLICE (p); + +if (time_slice <MIN_TIMESLICE ) + time_slice =MIN_TIMESLICE ; + +return time_slice; }/*
@@ -347,6 +375,7 @@if (sleep_time >0 ) {int sleep_avg; +int slice =task_timeslice (p);/* * This code gives a bonus to interactive tasks.
@@ -355,8 +384,15 @@* value here, based on ->last_run. The more time a task * spends sleeping, the higher the average gets - and the * higher the priority boost gets as well. +* +* Prevent tasks with an extremely high context switch +* rate from becoming CPU hogs by steadily inflating their +* sleep_avg until they starve legitimate sleepers. */ - sleep_avg = p->sleep_avg + sleep_time; +if (p->sleep_avg >MAX_SLEEP_AVG /2 ) + p->sleep_avg -=NEGATIVE_FEEDBACK (slice); + sleep_time %=MAX_ACCELERATION (slice); + sleep_avg = sleep_time + p->sleep_avg;/* * 'Overflow' bonus ticks go to the waker as well, so the
@@ -1176,10 +1212,8 @@* load-dependent, as the frequency of array switched decreases with * increasing number of running tasks: */ -#define EXPIRED_STARVING (rq) \ - (STARVATION_LIMIT && ((rq)->expired_timestamp && \ - (jiffies - (rq)->expired_timestamp >= \ - STARVATION_LIMIT * ((rq)->nr_running) + 1))) +#define EXPIRED_STARVING (rq) ((rq)->expired_timestamp && \ + time_after_eq(jiffies, (rq)->expired_timestamp+STARVATION_LIMIT))/* * This function gets called by the timer code, with HZ frequency.
@@ -1193,6 +1227,10 @@int cpu =smp_processor_id ();runqueue_t *rq =this_rq ();task_t *p = current; +int array_switch =EXPIRED_STARVING (rq); + +if (FORCE_SWITCH && !rq->expired_timestamp) + rq->expired_timestamp = jiffies - (STARVATION_LIMIT /2 );if (rcu_pending (cpu))rcu_check_callbacks (cpu, user_ticks);
@@ -1242,23 +1280,52 @@/* put it at the end of the queue: */ dequeue_task (p, rq->active); -enqueue_task (p, rq->active); +enqueue_task (p, array_switch ? rq->expired : rq->active); + }else if (array_switch) { +if (!rq->expired_timestamp) + rq->expired_timestamp = jiffies; +dequeue_task (p, rq->active); +enqueue_task (p, rq->expired); } goto out; } -if (!--p->time_slice) { +if (unlikely (!--p->time_slice)) {dequeue_task (p, rq->active);set_tsk_need_resched (p); p->prio =effective_prio (p); p->time_slice =task_timeslice (p); p->first_time_slice =0 ; -if (!TASK_INTERACTIVE (p) ||EXPIRED_STARVING (rq)) { +if (!TASK_INTERACTIVE (p) || array_switch) {if (!rq->expired_timestamp) rq->expired_timestamp = jiffies;enqueue_task (p, rq->expired); }else enqueue_task (p, rq->active); + goto out; + } +/* +* Prevent a too long timeslice allowing a task to monopolize +* the CPU. We do this by splitting up the timeslice into +* smaller pieces. +* +* Note: this does not mean the task's timeslices expire or +* get lost in any way, they just might be preempted by +* another task of equal priority. (one with higher +* priority would have preempted this task already.) We +* requeue this task to the end of the list on this priority +* level, which is in essence a round-robin of tasks with +* equal priority. +* +* If expired tasks are starving, switch arrays ASAP. +*/ +if (!(p->time_slice %TIMESLICE_GRANULARITY ) || array_switch) { +dequeue_task (p, rq->active); +set_tsk_need_resched (p); + p->prio =effective_prio (p); +enqueue_task (p, array_switch ? rq->expired : rq->active); +if (array_switch && !rq->expired_timestamp) + rq->expired_timestamp = jiffies; } out:spin_unlock (&rq->lock);
@@ -1297,8 +1364,13 @@ rq =this_rq ();release_kernel_lock (prev); +#if 0 // MIKEDIDIT prev->last_run = jiffies; +#endif spin_lock_irq (&rq->lock); +#if 1 // MIKEDIDIT + prev->last_run = jiffies; +#endif /* * if entering off of a kernel preemption go straight
@@ -1680,7 +1752,7 @@*/ int task_prio (task_t *p) { -return p->prio -MAX_USER_RT_PRIO ; +return p->prio -MAX_RT_PRIO ; }/**
kernel/sysctl.c +59 -1
84 lines not on this page · raw message
include/linux/sysctl.h +18 -1
@@ -66,7 +66,8 @@CTL_DEV =7 ,/* Devices */ CTL_BUS =8 ,/* Busses */ CTL_ABI =9 ,/* Binary emulation */ - CTL_CPU=10 /* CPU stuff (speed scaling, etc) */ +CTL_CPU =10 ,/* CPU stuff (speed scaling, etc) */ + CTL_SCHED=11, /* scheduler tunables */ };/* CTL_BUS names: */
@@ -157,6 +158,22 @@ VM_LOWER_ZONE_PROTECTION=20,/* Amount of protection of lower zones */ }; +/* Tunable scheduler parameters in /proc/sys/sched/ */ +enum { +SCHED_MIN_TIMESLICE =1 ,/* minimum process timeslice */ +SCHED_MAX_TIMESLICE =2 ,/* maximum process timeslice */ +SCHED_CHILD_PENALTY =3 ,/* penalty on fork to child */ +SCHED_PARENT_PENALTY =4 ,/* penalty on fork to parent */ +SCHED_EXIT_WEIGHT =5 ,/* penalty to parent of CPU hog child */ +SCHED_PRIO_BONUS_RATIO =6 ,/* percent of max prio given as bonus */ +SCHED_INTERACTIVE_DELTA =7 ,/* delta used to scale interactivity */ +SCHED_MAX_SLEEP_AVG =8 ,/* maximum sleep avg attainable */ +SCHED_STARVATION_LIMIT =9 ,/* no re-active if expired is starved */ +SCHED_NODE_THRESHOLD =10 ,/* NUMA node rebalance threshold */ +SCHED_MAX_ACCELERATION =11 ,/* maximum bonus slices per activation */ +SCHED_NEG_FEEDBACK =12 ,/* percent slice onus per activation */ +SCHED_FORCE_SWITCH =13 ,/* force switch to expired array */ +};/* CTL_NET names: */ enum
Quoted 1 line · Mike Galbraith
Anyway, it's attached if you want to play with it ;-)
P.S. it's been beaten hard on my little UP (+preempt, but) box. It is interesting to watch, and I haven't been able to piss it off yet.
Quoted 5 lines · Mike Galbraith
Greetings potential victims :)
Care to see if the attached cures your woes?
Quoted 7 lines
This is a mixture of Ingo's last posted plus the scheduler tuning knobs patch (
/proc/sys/sched/*). I added three new knobs to watch the effect on different loads.max_accel_sliceslimits the amount ofsleep_timeyou may add in one activation.retard_prct_slicesis a percentage of a slice to deduct fromsleep_timeeach activation (negative feedback for heavy context switchers.. dang irmanprocess_load).force_switchis there because I'm playing :) I didn't do much to the scheduler itself, only made it switch arrays in something closer to a square wave. With the settings as in the patch, and running a kernel build, top and irman, irman reports worst case response times of 150ms for NULL load, 316ms formemory_load, 414 forio_load, and 504ms forprocess_load.Anyway, it's attached if you want to play with it ;-)
-Mike
Oh, it's against virgin 2.5.66.
Thanks, running 2.5.66 with the patch applied I could no more reproduce the problem. I haven't tried playing with the knobs.
Michal
Quoted 10 lines · Michal Schmidt
Quoted 6 lines · Mike Galbraith
Greetings potential victims :)
Care to see if the attached cures your woes?
This is a mixture of Ingo's last posted plus the scheduler tuning knobs patch (/proc/sys/sched/*). I added three new knobs to watch the effect on different loads.max_accel_sliceslimits the amount ofsleep_timeyou may add in one activation.retard_prct_slicesis a percentage of a slice to deduct fromsleep_timeeach activation (negative feedback for heavy context switchers.. dang irmanprocess_load).force_switchis there because I'm playing :) I didn't do much to the scheduler itself, only made it switch arrays in something closer to a square wave. With the settings as in the patch, and running a kernel build, top and irman, irman reports worst case response times of 150ms for NULL load, 316ms formemory_load, 414 forio_load, and 504ms forprocess_load. Anyway, it's attached if you want to play with it ;-)
-Mike
Oh, it's against virgin 2.5.66.Thanks, running 2.5.66 with the patch applied I could no more reproduce the problem. I haven't tried playing with the knobs.
Thanks for the info.
-Mike
Quoted 2 lines · Andrew Ebling
I'm seeing similar on 2.5.66; xmms pauses when doing disk intensive tasks.
Just out of curiosity, do mpg123 (or mpg321) and ogg123 behave in the same manner? I have found that XMMS is prone to skips, pauses and hangs, but not so ogg123.
________________________________________________________________________
Felipe Alfaro Solana
Linux Registered User #287198
http://counter.li.org
Quoted 8 lines · Andrew Ebling
Quoted 3 lines · Michal Schmidt
I'm seeing serious interactivity problems in 2.5.65, 2.5.66, 2.5.65-mm4 and 2.5.64-bk2.
I couldn't reproduce it on 2.5.64, 2.5.64-bk1.I'm seeing similar on 2.5.66; xmms pauses when doing disk intensive tasks.
The only problem I see with X under 2.5.66 is that if you stop using it long enough to have the screensaver kick in you can never unlock it. This is in 2.5.6[56] and not in 2.5.59.
Having the screen locked certainly does hurt interactivity;-)
Signature · 3 lines
bill davidsen <[email protected]> CTO, TMR Associates, Inc Doing interesting things with little computers since 1979.