CPU issues with Speakup

CPU issues with Speakup

[Hermann]

Hi,
now that I've tested Speakup on several machines with several GRML 
versions, I found out that Speakup in general seems to use the CPU very 
extensively.
On my desktop machine with GRML 2008-11, I observe a CPU load of nearly 
100% when Speakup is active; the load is for sys, e.g. 99,1%sy.
The hardware is a 1 Ghz AMD Duron, 256 MB RAM. I use GRML in text mode only.
On my notebook I've only tested the latest GRML 2009-05-RC1 in live mode.
Although the CPU load is quite normal, e.g. about 6.0%sy, I observe 
about 12500 wakeups per second, which seems extremely high.
Hardware: Intel processor of 2,7 Ghz and 512 MB RAM.
In both cases i use Espeakup from GIT.
Is this known?
Hermann

Re: CPU issues with Speakup

[Kirk Reiser]

On Fri, 29 May 2009, Hermann wrote:

> On my notebook I've only tested the latest GRML 2009-05-RC1 in live mode.
> Although the CPU load is quite normal, e.g. about 6.0%sy, I observe about 
> 12500 wakeups per second, which seems extremely high.
> Hardware: Intel processor of 2,7 Ghz and 512 MB RAM.

That does seem pretty excessive.  Where are you getting that wakeup
rate from?  I don't think our scheduling should be bringing anywhere
near that figure.
   Kirk
--
Kirk Reiser				The Computer Braille Facility
e-mail: kirk@braille.uwo.ca		University of Western Ontario
phone: (519) 661-3061

Re: CPU issues with Speakup

[Hermann]

On 29.05.2009 17:32, Kirk Reiser wrote:
> On Fri, 29 May 2009, Hermann wrote:
>
>> On my notebook I've only tested the latest GRML 2009-05-RC1 in live mode.
>> Although the CPU load is quite normal, e.g. about 6.0%sy, I observe about
>> 12500 wakeups per second, which seems extremely high.
>> Hardware: Intel processor of 2,7 Ghz and 512 MB RAM.
>
> That does seem pretty excessive.  Where are you getting that wakeup
> rate from?  I don't think our scheduling should be bringing anywhere
> near that figure.
Above all from Speakup. I forgot the exact percentage, but Speakup was 
the top reason.
Hermann

Re: CPU issues with Speakup

[Samuel Thibault]

Hermann, le Fri 29 May 2009 17:37:43 +0200, a écrit :
> On 29.05.2009 17:32, Kirk Reiser wrote:
> >On Fri, 29 May 2009, Hermann wrote:
> >>On my notebook I've only tested the latest GRML 2009-05-RC1 in live mode.
> >>Although the CPU load is quite normal, e.g. about 6.0%sy, I observe about
> >>12500 wakeups per second, which seems extremely high.
> >>Hardware: Intel processor of 2,7 Ghz and 512 MB RAM.
> >
> >That does seem pretty excessive.  Where are you getting that wakeup
> >rate from?  I don't think our scheduling should be bringing anywhere
> >near that figure.
> Above all from Speakup. I forgot the exact percentage, but Speakup was 
> the top reason.

How are you reading that percentage?

Samuel

Re: CPU issues with Speakup

[Hermann]

On 29.05.2009 17:40, Samuel Thibault wrote:
> Hermann, le Fri 29 May 2009 17:37:43 +0200, a écrit :
>> On 29.05.2009 17:32, Kirk Reiser wrote:
>>> On Fri, 29 May 2009, Hermann wrote:
>>>> On my notebook I've only tested the latest GRML 2009-05-RC1 in live mode.
>>>> Although the CPU load is quite normal, e.g. about 6.0%sy, I observe about
>>>> 12500 wakeups per second, which seems extremely high.
>>>> Hardware: Intel processor of 2,7 Ghz and 512 MB RAM.
>>>
>>> That does seem pretty excessive.  Where are you getting that wakeup
>>> rate from?  I don't think our scheduling should be bringing anywhere
>>> near that figure.
>> Above all from Speakup. I forgot the exact percentage, but Speakup was
>> the top reason.
>
> How are you reading that percentage?
>
I use powertop.
Hermann

Re: CPU issues with Speakup

[Chris Brannon]

[-- Attachment #1: Type: text/plain, Size: 165 bytes --]

I am attaching the output from an execution of "powertop -d" on my
system.  During this run, espeakup had an average of 1669.1 wakeups / second.

Regards,
-- Chris


[-- Attachment #2: powertop-dump.txt --]
[-- Type: text/plain, Size: 1524 bytes --]

PowerTOP 1.11   (C) 2007, 2008 Intel Corporation 

Collecting data for 15 seconds 


< Detailed C-state information is not available.>
P-states (frequencies)
Wakeups-from-idle per second : 2075.9	interval: 15.0s
no ACPI power usage estimate available
Top causes for wakeups:
  80.4% (1669.1)          espeakup : hrtimer_start_range_ns (hrtimer_wakeup) 
  14.9% (309.2)       <interrupt> : VIA8237 
   2.5% ( 51.1)       <interrupt> : ra0 
   1.3% ( 26.3)     <kernel core> : hrtimer_start_range_ns (tick_sched_timer) 
   0.4% (  8.0)   <kernel module> : usb_hcd_poll_rh_status (rh_timer_func) 
   0.2% (  3.3)       <interrupt> : PS/2 keyboard/mouse/touchpad 
   0.1% (  2.0)     <kernel core> : clocksource_register (clocksource_watchdog) 
   0.1% (  1.5)             emacs : synth_printf (thread_wake_up) 
   0.0% (  1.0)            dhcpcd : MlmeInit (MlmePeriodicExec) 
   0.0% (  0.3)     <kernel core> : neigh_table_init_no_netlink (neigh_periodic_timer) 
   0.0% (  0.3)   <kernel module> : neigh_table_init_no_netlink (neigh_periodic_timer) 
   0.0% (  0.2)       <interrupt> : sata_via 
   0.0% (  0.2)              init : hrtimer_start_range_ns (hrtimer_wakeup) 
   0.0% (  0.2)     <kernel core> : page_writeback_init (wb_timer_fn) 
   0.0% (  0.1)             kinit : start_this_handle (commit_timeout) 
   0.0% (  0.1)          events/0 : queue_delayed_work (delayed_work_timer_fn) 
   0.0% (  0.1)               mpd : announce_edge (kd_nosound) 
   0.0% (  0.1)     <kernel core> : announce_edge (kd_nosound) 

Re: CPU issues with Speakup

[William Hubbs]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On Fri, May 29, 2009 at 11:12:39AM -0500, Chris Brannon wrote:
> I am attaching the output from an execution of "powertop -d" on my
> system.  During this run, espeakup had an average of 1669.1 wakeups / second.
 
Ok, I know what is happening here.  Espeakup is using a select() call to
read the softsynth with a timeout of 1 microsecond, so that would be
about right.  The issue is with espeakup, not speakup.  I will take a
look at how I am using the select call.

> PowerTOP 1.11   (C) 2007, 2008 Intel Corporation 
> 
> Collecting data for 15 seconds 
> 
> 
> < Detailed C-state information is not available.>
> P-states (frequencies)
> Wakeups-from-idle per second : 2075.9	interval: 15.0s
> no ACPI power usage estimate available
> Top causes for wakeups:
>   80.4% (1669.1)          espeakup : hrtimer_start_range_ns (hrtimer_wakeup) 
>   14.9% (309.2)       <interrupt> : VIA8237 
>    2.5% ( 51.1)       <interrupt> : ra0 
>    1.3% ( 26.3)     <kernel core> : hrtimer_start_range_ns (tick_sched_timer) 
>    0.4% (  8.0)   <kernel module> : usb_hcd_poll_rh_status (rh_timer_func) 
>    0.2% (  3.3)       <interrupt> : PS/2 keyboard/mouse/touchpad 
>    0.1% (  2.0)     <kernel core> : clocksource_register (clocksource_watchdog) 
>    0.1% (  1.5)             emacs : synth_printf (thread_wake_up) 
>    0.0% (  1.0)            dhcpcd : MlmeInit (MlmePeriodicExec) 
>    0.0% (  0.3)     <kernel core> : neigh_table_init_no_netlink (neigh_periodic_timer) 
>    0.0% (  0.3)   <kernel module> : neigh_table_init_no_netlink (neigh_periodic_timer) 
>    0.0% (  0.2)       <interrupt> : sata_via 
>    0.0% (  0.2)              init : hrtimer_start_range_ns (hrtimer_wakeup) 
>    0.0% (  0.2)     <kernel core> : page_writeback_init (wb_timer_fn) 
>    0.0% (  0.1)             kinit : start_this_handle (commit_timeout) 
>    0.0% (  0.1)          events/0 : queue_delayed_work (delayed_work_timer_fn) 
>    0.0% (  0.1)               mpd : announce_edge (kd_nosound) 
>    0.0% (  0.1)     <kernel core> : announce_edge (kd_nosound) 
> 

> _______________________________________________
> Speakup mailing list
> Speakup@braille.uwo.ca
> http://speech.braille.uwo.ca/mailman/listinfo/speakup

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.11 (GNU/Linux)

iEYEARECAAYFAkogFQ0ACgkQblQW9DDEZTh2hwCghbU6ciU8uU8gyqxf96VRkIWC
GKMAn1TKtI5ebj4R1YGs6A/itNHMqYUV
=/8v2
-----END PGP SIGNATURE-----

Re: CPU issues with Speakup

[Kerry Hoath]

I know for a fact espekupd wakes up the processer 100 times a second 
according to powertop.
Use of usleep perhaps? Haven't looked at the code.
raises wakeup rate on my eeepc from 12.3 a second to 112.3 once espeakupd 
starts.
Regards, Kerry.

----- Original Message ----- 
From: "Kirk Reiser" <kirk@braille.uwo.ca>
To: "Speakup is a screen review system for Linux." <speakup@braille.uwo.ca>
Sent: Friday, May 29, 2009 11:32 PM
Subject: Re: CPU issues with Speakup


On Fri, 29 May 2009, Hermann wrote:

> On my notebook I've only tested the latest GRML 2009-05-RC1 in live mode.
> Although the CPU load is quite normal, e.g. about 6.0%sy, I observe about
> 12500 wakeups per second, which seems extremely high.
> Hardware: Intel processor of 2,7 Ghz and 512 MB RAM.

That does seem pretty excessive.  Where are you getting that wakeup
rate from?  I don't think our scheduling should be bringing anywhere
near that figure.
   Kirk
--
Kirk Reiser The Computer Braille Facility
e-mail: kirk@braille.uwo.ca University of Western Ontario
phone: (519) 661-3061
_______________________________________________
Speakup mailing list
Speakup@braille.uwo.ca
http://speech.braille.uwo.ca/mailman/listinfo/speakup

Re: CPU issues with Speakup

[Kirk Reiser]

Right.  We've discussed this on irc and Chris Brannon and William have
pushed a couple patches to espeakup to bring the wakeups way down.  I
encourrage folks to upgrade from git if you're experiencing load
problems from speakup/espeakup.
   Kirk

--
Kirk Reiser				The Computer Braille Facility
e-mail: kirk@braille.uwo.ca		University of Western Ontario
phone: (519) 661-3061