Re: LCDd Command/Display Latency

On 14/06/12 19:20, Caylan Van Larson wrote:
> On Jun 14, 2012, at 9:23 AM, Martin T Jones wrote:
>
>> On 13/06/12 22:59, Caylan Van Larson wrote:
>>>
>>> The speed I'm seeing with libusb-1.0 is definitely faster.  One out 
>>> of every 20 commands I send are displayed immediately.  However, 
>>> most of the time the display lags about .5 seconds, up to one second 
>>> later.  This lack of real-time response makes a user-interface 
>>> difficult to use.  I've tested my code and it is definitely in the 
>>> display (not in the key press).

>>> Is there a way to tighten the main.c loop by way of main.h 
>>> constants?  I understand that tightening the loop is at the expense 
>>> of CPU cycles... how fast can I reasonably push it?
>>>
>>>> /* You should be able to modify the following freqencies... */
>>>> #define RENDER_FREQ 8
>>>> /* We want 8 frames per second */
>>>> #define PROCESS_FREQ 32
>>>> /* And 32 times per second processing of messages and keypresses. */

>> My system is complied with a higher process frequency (#define 
>> PROCESS_FREQ 100) which is needed to handle the IR receiver data in 
>> time but I don't remember a problem at the standard value.
>>
>> I've just done a quick test and the slowest (client_set -name 
>> example) is just over 30ms.
>>
>> martin@Ace:~$ socat -vlu STDIO,ignoreeof TCP4:LiFi.local:13666 
>> 2>picoLCD.log
>>
>> > 2012/06/14 14:45:54.881604  length=6 from=0 to=5
>> hello
>> < 2012/06/14 14:45:54.889096  length=73 from=0 to=72
>> connect LCDproc 0.5dev protocol 0.3 lcd wid 20 hgt 2 cellwid 5 cellhgt 8
>> > 2012/06/14 14:46:21.853583  length=25 from=6 to=30
>> client_set -name example
>> < 2012/06/14 14:46:21.884362  length=8 from=73 to=80
>> success
>> > 2012/06/14 14:46:32.133078  length=19 from=31 to=49
>> screen_add example
>> < 2012/06/14 14:46:32.137076  length=8 from=81 to=88
>> success
>> > 2012/06/14 14:46:47.861075  length=27 from=50 to=76
>> client_add_key  -shared F1
>> < 2012/06/14 14:46:47.864711  length=8 from=89 to=96
>> success
>> > 2012/06/14 14:47:12.741589  length=38 from=77 to=114
>> widget_add example exampleTitle title
>> < 2012/06/14 14:47:12.747946  length=8 from=97 to=104
>> success
>> > 2012/06/14 14:47:20.717590  length=48 from=115 to=162
>> widget_set example exampleTitle "Example Title"
>> < 2012/06/14 14:47:20.736064  length=8 from=105 to=112
>> success
>> > 2012/06/14 14:47:46.509084  length=27 from=163 to=189
>> widget_add example i1 icon
>> < 2012/06/14 14:47:46.522206  length=8 from=113 to=120
>> success
>> > 2012/06/14 14:47:58.929076  length=31 from=190 to=220
>> widget_set example i1 1 2 prev
>> < 2012/06/14 14:47:58.933163  length=8 from=121 to=128
>> success
>> < 2012/06/14 14:48:15.196001  length=15 from=129 to=143
>> listen example
>> > 2012/06/14 14:53:16.309597  length=32 from=221 to=252
>> widget_add example line1 string
>> < 2012/06/14 14:53:16.314173  length=8 from=144 to=151
>> success
>> > 2012/06/14 14:53:26.605096  length=32 from=253 to=284
>> widget_add example line2 string
>> < 2012/06/14 14:53:26.634674  length=8 from=152 to=159
>> success
>> > 2012/06/14 14:53:33.693094  length=46 from=285 to=330
>> widget_set example line1 1 1 "Testing Line 1"
>> < 2012/06/14 14:53:33.696917  length=8 from=160 to=167
>> success
>> > 2012/06/14 14:53:39.429598  length=46 from=331 to=376
>> widget_set example line2 1 2 "Testing Line 2"
>> < 2012/06/14 14:53:39.436049  length=8 from=168 to=175
>> success
>
> More debugging from my end.
>
> < 2012/06/14 13:19:39.104789  length=7 from=336 to=342
> key F1
> > 2012/06/14 13:19:40.105947  length=146 from=993 to=1138
> widget_set picolcd line1 1 1 "CoPOS - Status      "
> widget_set picolcd line2 1 2 "Up/Down to Choose >"
> screen_set picolcd -backlight on
> output 3
> < 2012/06/14 13:19:40.109371  length=32 from=343 to=374
> success
> success
> success
> success
>
> Hrmm... so maybe not socat?
>
> Caylan

I've had a more detailed look at the LCDd response times on my system 
and they are mostly random values up to 10ms, as one would expect with 
100Hz process frequency (#define PROCESS_FREQ 100).

The problem is the design of do_mainloop(void) in server/main.c.

The sockets are polled in sock_poll_clients(); (select() with zero 
time-out), the required processing is performed then LCDd *sleeps* until 
the next processing slot (usleep(sleeptime);).

Ideally there should be a single select() with a time-out appropriate 
for the time based processing (polling of some keypads and scrolling of 
those widgets that support it), all other processing should be as 
required. Unfortunately that is not a trivial change.

Martin

_______________________________________________
LCDproc mailing list
[email protected]
http://lists.omnipotent.net/mailman/listinfo/lcdproc