[Opensim-users] Slow LSL timers

Mon Dec 6 07:31:37 UTC 2010

Hi

If I remember it correctly, and mind you I'm not up to date on the latest
code, the reason for one central timer is for fair scheduling.
If the script can schedule any amount of timers then it could end up eating
multiple threads, even if invoking the scripts "dedicated" thread. Scripts
don't have dedicated threads either.

So by logic the fair way to handle this is to queue timers to be executed in
a queue for each script (or a global queue).

You can easily bypass this by calling System.*.Timer from your script just
like in any .Net-app, but it requires some lowering of permissions I guess.
Not sure about the details. It wouldn't be hard to implement an OSSL
function for this.

Br,
 Tedd

-----Original Message-----
From: opensim-users-bounces at lists.berlios.de
[mailto:opensim-users-bounces at lists.berlios.de] On Behalf Of Jorge Lima
Sent: 6. desember 2010 02:46
To: opensim-users at lists.berlios.de
Subject: Re: [Opensim-users] Slow LSL timers

Thank you Paul, I will look into Ptolemy.
I briefly looked at the code for timers and it seems a bit strange to me.

>From what I know of SL I expected each script to be little more than a
self-contained thread or process that mapped itself as thinly as
possible to the underlying physics engine and operating system.

So I don't understand why we need a centralized "timer manager".
Couldn't the underlying OS be managing the timers on its own - exactly
as if I had called setitimer in C, and received a SIGALRM, etc. ?

On Mon, Dec 6, 2010 at 12:54 AM, Paul Fishwick <fishwick at cise.ufl.edu>
wrote:
> I was going to wait until this week to show this link since it is not
> completely ready (yet), but we have an implementation using
> a region module for simple harmonic motion. We drive this motion
> through a separate package (Ptolemy) that I have used in my
> systems simulation class:
>
> http://www.cise.ufl.edu/~zezzell/ptolemy/ptolemy_opensim.html
>
> Ultimately, for several reasons including artificial delays, LSL
> seems inadequate for near real-time simulation. However, access
> to the raw timing facility in OpenSim is adequate.
>
> -p
>
> On 12/5/2010 7:30 PM, Jorge Lima wrote:
>>
>> My project is simple enough - something like throwing a ball off a
>> cliff with a horizontal impulse and recording its speed. It's for high
>> school physics teachers/students.
>>
>> It seems to me that I can either have a fast timer that simulates the
>> movement and generates points, or I can have a fast timer that trusts
>> the physics engine to provide an accurate simulation and records its
>> output :)
>>
>> I will look into the Timer.cs file and try out C# scripting tomorrow,
>> then get back to you. Thanks for the help!
>>
>> On Sat, Dec 4, 2010 at 6:23 AM, dz<dz at bitzend.net>  wrote:
>>>
>>> I'd have to agree with Justin...
>>>
>>> Expecting event handling to be optimized is too optimistic, but  there
>>> are
>>> alternatives.  The .Net/Mono libraries already include timer functions
>>> that
>>> support fast and accurate timing.
>>>
>>> Relying on an LSL event trigger to happen every 4 frames MAY work in SL
>>> but
>>> it is a design begging to cause lag and  stop itself.  Designs like that
>>> are
>>> also susceptible to event buffer overflows  where your events are
"lost",
>>> leaving you to wonder why your "reliable" script  sometimes just isn't..
>>> The ONLY LSL events I have seen  requiring sub-second repeats  would be
>>> in
>>> an AO, but timer based AO's are obsoleted in OpenSim by the use of the
>>> Animation_Changed event.
>>>
>>> In the extreme, if you are trying to be  very exact,,  you would want to
>>> insure the timings were made by processes independent of the region time
>>> dialations, not the ones that may be responsible for them.
>>>
>>>
>>>
>>> On Fri, Dec 3, 2010 at 5:35 PM,<opensim-users-request at lists.berlios.de>
>>> wrote:
>>>>
>>>> ---------- Forwarded message ----------
>>>> From: Jorge Lima<jorge.manuel at gmail.com>
>>>>
>>>> Subject: [Opensim-users] Slow LSL timers
>>>> Ran 100 timer events with a single if(++i>  100) inside. I suppose
>>>> decrementing and comparing to 0 might be faster, but I also
>>>> copy-pasted 100 inlined if clauses and the time to run them was
>>>> negligible.
>>>>
>>>> llSetTimerEvent(0.1) ->  59 s ->  490% overhead
>>>> llSetTimerEvent(0.2) ->  59 s ->  195% overhead
>>>> llSetTimerEvent(0.5) ->  59 s ->  18% overhead
>>>> llSetTimerEvent(0.75) ->  81 s ->  8% overhead
>>>> llSetTimerEvent(1.0) ->  108 s ->  8% overhead
>>>> llSetTimerEvent(2.0) ->  209 s ->  4.5% overhead
>>>> llSetTimerEvent(4.0) ->  407 s ->  1.75% overhead
>>>>
>>>> I am guessing that the overhead is, therefore, internal to the timer
>>>> implementation. I haven't tested raw C# timers as comparison but 100ms
>>>> is an eternity in modern computing, I can ping servers across the
>>>> Atlantic ocean in less than that!
>>>>
>>>> I went through the code but I haven't yet found the part that handles
>>>> timers. It's not in EventManager.
>>>> Does anyone know what's going on? The code works fine in SL.
>>>> Should I be posting this in the dev list? I'd like to see if I can fix
>>>> it.
>>>> Any thoughts on how to get fast timers in OpenSim?
>>>> Faster computer? (I did check and was only getting about 20% CPU
>>>> usage...)
>>>> Porting them to a Real-Time operating system? :)
>>>>
>>>>
>>>>
>>>> ---------- Forwarded message ----------
>>>> From: Justin Clark-Casey<jjustincc at googlemail.com>
>>>> To: opensim-users at lists.berlios.de
>>>> Date: Sat, 04 Dec 2010 01:35:32 +0000
>>>> Subject: Re: [Opensim-users] Slow LSL timers
>>>
>>>> I'm not familiar with the this area of the code but I had a quick look
>>>> since I was curious (the meat is in
>>>> OpenSim/Region/ScriptEngine/Shared/Api/Plugins/Timer.cs).  The code
>>>> isn't
>>>> using C# timers so I'm not totally surprised at the behaviour that
>>>> you're
>>>> seeing.
>>>>
>>>> I suspect addressing this would require a considerably different
>>>> implementation.
>>>>
>>> _______________________________________________
>>> Opensim-users mailing list
>>> Opensim-users at lists.berlios.de
>>> https://lists.berlios.de/mailman/listinfo/opensim-users
>>>
>>>
>> _______________________________________________
>> Opensim-users mailing list
>> Opensim-users at lists.berlios.de
>> https://lists.berlios.de/mailman/listinfo/opensim-users
>>
>
>
> --
> Paul Fishwick, PhD
> Professor and Director, Digital Arts and Sciences
> CISE Department, CSE 301
> University of Florida
> Gainesville, FL 32611
> Email: fishwick at cise.ufl.edu
> Web: http://www.cise.ufl.edu/~fishwick
> Blog: http://www.representationz.com
>
> _______________________________________________
> Opensim-users mailing list
> Opensim-users at lists.berlios.de
> https://lists.berlios.de/mailman/listinfo/opensim-users
>
_______________________________________________
Opensim-users mailing list
Opensim-users at lists.berlios.de
https://lists.berlios.de/mailman/listinfo/opensim-users