POV-Ray : Newsgroups : povray.animations : Memory Leak During Animations (v3.7.0 rc6, win7) Server Time
29 Mar 2024 07:43:52 EDT (-0400)
  Memory Leak During Animations (v3.7.0 rc6, win7) (Message 11 to 20 of 20)  
<<< Previous 10 Messages Goto Initial 10 Messages
From: Christian Froeschlin
Subject: Re: Memory Leak During Animations (v3.7.0 rc6, win7)
Date: 3 Dec 2012 20:16:46
Message: <50bd4efe$1@news.povray.org>
Kenneth wrote:

> OBSERVATIONS:

you may be overcomplicating the situation a bit. It appears that

1. The first frame rendered determines whether to use bounding
    for all frames (this is a bug or design flaw, IMHO)

2. If bounding is disabled, frames with many objects render slowly
    (this is to be expected)

There is nothing magical about "adding" objects between frames.
An animition is almost rendered as if each frame were a separate
scene, with some automic values for variables like clock, special
output file names, and apparently an unfortunate persistence of
the predetermined bounding setting.


Post a reply to this message

From: Sereib
Subject: Re: Memory Leak During Animations (v3.7.0 rc6, win7)
Date: 4 Dec 2012 12:00:01
Message: <web.50be2bcff8006e78fee2bcbd0@news.povray.org>
Dear all,

in my animation, the number of objects kept absolutely constant from the first
to the final frame: Approx. 1 million rocks orbiting a planet. The thing which
changed (beside the position of the rocks) is the nature of the rocks: A certain
fraction (those far away from the camera location) are just spheres, while the
rocks closer to the camera are isosurfaces. Since the camera comes closer and
closer to the planet, the fraction of rocks consisting of isosurfaces increases
by the same number the fraction of spheres decreases.

Maybe this help?

When I've some more time, I try to check how far I can simplify the scene but
still observe the memory leak, I'll let you know!


Post a reply to this message

From: Le Forgeron
Subject: Re: Memory Leak During Animations (v3.7.0 rc6, win7)
Date: 4 Dec 2012 13:19:46
Message: <50be3ec2$1@news.povray.org>
Le 04/12/2012 02:16, Christian Froeschlin nous fit lire :
> Kenneth wrote:
> 
>> OBSERVATIONS:
> 
> you may be overcomplicating the situation a bit. It appears that
> 
> 1. The first frame rendered determines whether to use bounding
>    for all frames (this is a bug or design flaw, IMHO)

I agree. Sound wrong. Make sense with the observations so far.

> 
> 2. If bounding is disabled, frames with many objects render slowly
>    (this is to be expected)
> 
> There is nothing magical about "adding" objects between frames.
> An animition is almost rendered as if each frame were a separate
> scene, with some automic values for variables like clock, special
> output file names, and apparently an unfortunate persistence of
> the predetermined bounding setting.

From previous experiment (by Kenneth), I would conclude so far that
sphere does not seem to leak.

And from Sereib:
> Dear all,
> 
> in my animation, the number of objects kept absolutely constant from the first
> to the final frame: Approx. 1 million rocks orbiting a planet. The thing which
> changed (beside the position of the rocks) is the nature of the rocks: A certain
> fraction (those far away from the camera location) are just spheres, while the
> rocks closer to the camera are isosurfaces. Since the camera comes closer and
> closer to the planet, the fraction of rocks consisting of isosurfaces increases
> by the same number the fraction of spheres decreases.

Now, there is another candidate for leak: isosurface.
Maybe a bit more details about the used isosurface (was precompute used
? other setting(s) ?)


Post a reply to this message

From: Kenneth
Subject: Re: Memory Leak During Animations (v3.7.0 rc6, win7)
Date: 4 Dec 2012 17:30:01
Message: <web.50be7880f8006e78c2d977c20@news.povray.org>
Christian Froeschlin <chr### [at] chrfrde> wrote:
> Kenneth wrote:
>
> > OBSERVATIONS:

> 2. If bounding is disabled, frames with many objects render slowly
>     (this is to be expected)

The problem I perceive is that there's some kind of 'extra' slow-down happening
--in 3.6.2--, not necessarily connected with the expected one of simply adding
more objects. The rendering time of my own test scene really does slow to a
*crawl*, and only after about 50 or so new frames (new simple spheres.) The
thing is, when the scene runs as it's supposed to (i.e., with Bounding_Threshold
set to 0 or using one of the other useful tricks), the rendering goes very fast,
and there's no perceived slowdown at all from frame to frame, even after
hundreds of new spheres have been added.

Perhaps it *is* some kind of (hidden) memory leak (the results seem to indicate
it.) Just a guess, though.


Post a reply to this message

From: Christian Froeschlin
Subject: Re: Memory Leak During Animations (v3.7.0 rc6, win7)
Date: 4 Dec 2012 19:05:15
Message: <50be8fbb$1@news.povray.org>
Kenneth wrote:

> The problem I perceive is that there's some kind of 'extra' slow-down happening
> --in 3.6.2--, not necessarily connected with the expected one of simply adding
> more objects.

Try this:

1. In your existing experiment, take note of one of the later frames
that render very slowly with some number N of objects and the time used
to render that frame

2. Create a two-frame animation
a. First frame dummy with 1 object to force the problem
b. Second frame with the same number of objects N as the test frame.

If the time used for the second frame in 2. is roughly the
same as the original test frame there is no accumlated slowdown
or resource leak. Instead you are simply observing that, without
bounding, render time is very high when object count is high.


Post a reply to this message

From: Kenneth
Subject: Re: Memory Leak During Animations (v3.7.0 rc6, win7)
Date: 4 Dec 2012 20:40:00
Message: <web.50bea4baf8006e78c2d977c20@news.povray.org>
Christian Froeschlin <chr### [at] chrfrde> wrote:

> 2. Create a two-frame animation
> a. First frame dummy with 1 object to force the problem
> b. Second frame with the same number of objects N as the test frame.

Ah, a clever little experiment. I'll try it later and report back...

>
> ...Instead you are simply observing that, without
> bounding, render time is very high when object count is high.

Hmm, I didn't know that. I assumed that a 'normal' slowdown would just be the
(miniscule) accumulation of the 900-byte-per-sphere memory increases per frame,
irrespective of Bounding working or not.  Didn't realize how *important*
Bounding is to a faster render!


Post a reply to this message

From: clipka
Subject: Re: Memory Leak During Animations (v3.7.0 rc6, win7)
Date: 5 Dec 2012 08:03:42
Message: <50bf462e$1@news.povray.org>
Am 05.12.2012 02:34, schrieb Kenneth:

>> ...Instead you are simply observing that, without
>> bounding, render time is very high when object count is high.
>
> Hmm, I didn't know that. I assumed that a 'normal' slowdown would just be the
> (miniscule) accumulation of the 900-byte-per-sphere memory increases per frame,
> irrespective of Bounding working or not.  Didn't realize how *important*
> Bounding is to a faster render!

Memory consumption (as such) has no significant impact on render time, 
as long as you don't run out of physical memory. (And if you do, your 
render will typically die a sudden death, and drag your whole system 
down to swap hell if you're running Windows.)

Then again, high memory consumption is usually caused by high object 
count, and the object count /does/ affect render time.

If it wasn't for bounding, render time would be roughly proportional to 
something between the object count N (in very sparsely populated scenes, 
or if you're using shadowless lights) and N^2 (in densely populated 
scenes with shadows). Bounding should cut that down to something between 
c*log(N) and (c*log(N))^2, with some constant c. For large scenes that's 
an epic win, almost irregardless of the factor c.


Post a reply to this message

From: Kenneth
Subject: Re: Memory Leak During Animations (v3.7.0 rc6, win7)
Date: 6 Dec 2012 22:35:00
Message: <web.50c16301f8006e78c2d977c20@news.povray.org>
"Kenneth" <kdw### [at] gmailcom> wrote:
> Christian Froeschlin <chr### [at] chrfrde> wrote:
>
> > 2. Create a two-frame animation
> > a. First frame dummy with 1 object to force the problem
> > b. Second frame with the same number of objects N as the test frame.
>
> Ah, a clever little experiment. I'll try it later and report back...

Here's the result (which agrees with what you've mentioned):

I first ran 3000 frames of animation (3000 total spheres at the end), with
Bounding OFF. Render time for the 3000-th frame: 102.16 seconds

Then ran 2-frame animation as you indicated, with Bounding OFF as well. Render
time for 2nd frame: 102.31 seconds.

That's pretty close!!

(I cheated a bit on the set-up--which didn't alter the results--to save some
tedium: For the 3000-frame animation, I just set Bounding_Threshold to
3003--effectively off. Then ran only frame 3000.)

BTW, as a reference to these render times: When Bounding is ON and working,
those 3000 spheres take just .5 sec to render! About 200X faster. Hmm...Bounding
is a GOOD thing.  ;-)


Post a reply to this message

From: Kenneth
Subject: Re: Memory Leak During Animations (v3.7.0 rc6, win7)
Date: 6 Dec 2012 22:45:00
Message: <web.50c16571f8006e78c2d977c20@news.povray.org>
P.S.  I apologize to the original poster; this Bounding_Threshold/slowdown thing
seems unrelated to his problem, as it turns out. Didn't mean to 'hi-jack' his
post.


Post a reply to this message

From: Kenneth
Subject: Re: Memory Leak During Animations (v3.7.0 rc6, win7)
Date: 7 Dec 2012 15:35:00
Message: <web.50c25198f8006e78c2d977c20@news.povray.org>
"Kenneth" <kdw### [at] gmailcom> wrote:
> "Kenneth" <kdw### [at] gmailcom> wrote:
> > Christian Froeschlin <chr### [at] chrfrde> wrote:
> >
> > > 2. Create a two-frame animation
> > > a. First frame dummy with 1 object to force the problem
> > > b. Second frame with the same number of objects N as the test frame.
> >
> > Ah, a clever little experiment. I'll try it later and report back...
>
> Here's the result...

I just realized that my test wasn't really valid for determining a memory-leak;
I should have run the entire 3000-frame animation (no cheating) to let any
possible problems *accumulate*. But I'm content to believe that a memory-leak
didn't happen.


Post a reply to this message

<<< Previous 10 Messages Goto Initial 10 Messages

Copyright 2003-2023 Persistence of Vision Raytracer Pty. Ltd.