 |
|
|
|
|
|
| |
| |
|
|
|
|
| |
| |
|
|
"Tony[B]" wrote:
>
> I appreciate your considering POV, but I wouldn't use this. It's way too
> dangerous, and would take too much of my resources.
On the point of the ressources, you should keep in mind that the machine
is used only when there are no interactive activity (screensaver) or
when the CPU is idle ( an activity under a low treshold ).
--
Gilles Equipe Architectures Paralleles Tel 33 01-69-15-42-25
FEDAK LRI bat 490 Fax 33 01-69-15-65-86
------ Universite Paris-Sud email: fed### [at] lri fr
XtremWeb F-91405 ORSAY Cedex http://www.lri.fr/~fedak
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
> With 480 clients out there on
> the net, I could have my image in a half hour instead of 3 days...
That's the point.
> The world is full of deviants, and it looks like a lot of them
> own computers now.
C'est la vie :)
> Also, denial attacks will be common -- particularly in the case of something
> like POV, it wouldn't take much effort to submit a bunch of files that would
> effectively prevent anybody from ever getting anything else processed. The
> script I mentioned earlier takes about two hours to render as just a 160x100
> thumbnail and lengthy rendering times definitely wasn't my goal...
Yes, an idea was to calculate few pixels, that could serve both as
predicting the rendering time of the scene, and as certificate for
correctness of the computations.
BTW I have one question about the way you distribute the rendering on
your different PCs. How does the distribution handles the various
frequency of the processors. I mean how the load is distributed to the
processors, statically or dynimically ?
--
Gilles Equipe Architectures Paralleles Tel 33 01-69-15-42-25
FEDAK LRI bat 490 Fax 33 01-69-15-65-86
------ Universite Paris-Sud email: fed### [at] lrifr
XtremWeb F-91405 ORSAY Cedex http://www.lri.fr/~fedak
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
>
> But thats the kicker.. I can then have a script that opens any
> file, and can read and write from/to it. I could make a .rhosts
> entry.. :-(
I mean that you don't have access to files outside of a temporary
directory.
--
Gilles Equipe Architectures Paralleles Tel 33 01-69-15-42-25
FEDAK LRI bat 490 Fax 33 01-69-15-65-86
------ Universite Paris-Sud email: fed### [at] lrifr
XtremWeb F-91405 ORSAY Cedex http://www.lri.fr/~fedak
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
Gilles Fedak wrote:
> Idea is to use the idle time of computers to render scenes provided by
> any one.
> Xtremweb allows the distribution of applications and the computing of
> tasks without to much effort, and there are clients available for Linux
> and Windows.
> Povray is for us a good test case for the platform.
> So what do you think about the usefulness of such a project ?
While I've no doubt about Povray being a good test case for this, the main
problem is about the usefulness of it for end users - a least for those who
create stills. Scenes that render slowly (let's say below 10 pps - roughly
an overnight render at 800*600) and that could benefit from it are often
(but not always) ones that also use large amounts of memory and disc space,
which would make their distribution uneasy. While there's usually no formal
relationship between render speed and memory/storage space requirements,
there's a practical one which is the price to pay for scene complexity :
several reflective/refractive objects in a scene involving high-quality
radiosity will make the scene crawl and the memory use rise.
Typically, people who have to meet deadlines for certain raytracing
competitions would be in cases like this ;-)
G.
--
**********************
http://www.oyonale.com
**********************
Graphic experiments
Pov-ray gallery
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
"Gilles Fedak" <fed### [at] lrifr> wrote in message
news:3AAA1D58.311A36C5@lri.fr...
> > With 480 clients out there on
> > the net, I could have my image in a half hour instead of 3 days...
> That's the point.
I didn't gather from your original explanation that you intended to break up
the render, it sounded more like you were talking about farming out
animation frame rendering. But if we agree, hey, great!
> Yes, an idea was to calculate few pixels, that could serve both as
> predicting the rendering time of the scene, and as certificate for
> correctness of the computations.
I see two problems with that plan as a test of fitness. First, if they can
figure out where you do the test renders, they can *easily* make that area
render quickly -- maybe even if you use multiple areas (for example, by
carefully placing bounding boxes). The second problem is the converse -- I
think it would be easy to make the small-area test-render take a very, very
long time. For example, layers of transparency, lots of lights, fog, high
max trace level, etc. Then I slam a few of those into your server and the
test-renders shut it down.
And trust me, I hate being the devil's advocate on this. I would love to
have distributed rendering...
> BTW I have one question about the way you distribute the rendering on
> your different PCs. How does the distribution handles the various
> frequency of the processors. I mean how the load is distributed to the
> processors, statically or dynimically ?
My distribution method is a little something I call "doing it by hand". :)
There isn't anything automated about how I do it now (unfortunately), other
than copying files around on the network and using remote control software
to kick off rendering on the other machines. I am toying with a couple ideas
but I haven't actually written any code yet.
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
On 2001-03-10 03:06, Margus Ramst <mar### [at] peakeduee> wrote:
>Gilles Fedak wrote:
>>
>> For the i/o point of view, we can easly restrict the size of the input
>> file to few Ko :)
>>
>
>Restricting the size of the input file is not enough, because even
>a tiny input file can easily generate a massive scene if it uses
>iterative elements such as loops and recursive macros. Disallowing such
>elements is probably not an option, because many people rely on them
>quite heavily.
You would have to put restrictions on the process. Most Unixes offer
quite a few "resource limits" which you can set. For distributed povray
the most important are IMHO:
RLIMIT_CPU
(limits the total CPU usage for one job)
RLIMIT_DATA, RLIMIT_STACK, RLIMIT_CORE
(limits the memory usage of a single process not all unixes have all
of these)
RLIMIT_NPROC
(limits the number of processes per user)
RLIMIT_NOFILE
(limits the number of open files per process)
Together with quotas, a chrooted environment and a positive nice factor,
the povray process(es) would be restricted from disturbing their host
computer.
What is sorely missing in standard Unix is a way to restrict network
connections: There are some approaches to that problem, but nothing
wide-spread yet (I'm guessing that a capability-based system will win
out, though). As long as a locally installed version of povray is run,
this isn't a problem, though, since povray doesn't talk across the net.
>As far as I can see, running a parse test on the server is the only
>reliable way of evaluating a submitted scene. The test sould check that
>both parse time and (more importantly IMO) memory usage are within
>acceptable limits. Of course with animations, a test parse must be
>performed for every frame, because different clock values can execute
>different branches in the script.
You can leave this to the slave servers: They can check whether there is
some output produces within a given time. If there isn't, they'll kill
the process and report failure.
hp
--
_ | Peter J. Holzer | All Linux applications run on Solaris,
|_|_) | Sysadmin WSR | which is our implementation of Linux.
| | | hjp### [at] wsracat |
__/ | http://www.hjp.at/ | -- Scott McNealy, Dec. 2000
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
On 2001-03-09 15:41, Gilles Fedak <fed### [at] lrifr> wrote:
>We plan to release a "peer to peer" Povray using the Xtremweb
>(http://xtremweb.net) global computing Platform.
I have downloaded the program, but I couldn't find much information on
how it works, and - most importantly - no source code on the site.
I don't like to run servers on my machines unless I have a clear idea of
what they do - of course some are just too large to audit completely
even if the source code is available, but the small size of the .jar and
.so make this entirely feasible.
hp
--
_ | Peter J. Holzer | All Linux applications run on Solaris,
|_|_) | Sysadmin WSR | which is our implementation of Linux.
| | | hjp### [at] wsracat |
__/ | http://www.hjp.at/ | -- Scott McNealy, Dec. 2000
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
"Gilles Tran" <tra### [at] inapginrafr> wrote in message
> While I've no doubt about Povray being a good test case for this, the main
> problem is about the usefulness of it for end users - a least for those
who
> create stills. Scenes that render slowly (let's say below 10 pps - roughly
> an overnight render at 800*600) and that could benefit from it are often
> (but not always) ones that also use large amounts of memory and disc
space,
> which would make their distribution uneasy.
For scenes that are reasonable in terms of memory use, images, etc,
non-radiosity -- eliminating all the major issues - then the next thing to
have to develop a user interest would be to let the end users view the
renderings that their machines are computing.
This way when you return to your computer you can see something partially
rendered and link back to a centrally controlled website that has stats on
all the parts of the scene, the author, originating IP of the job, and so
on.
In this way a community could be built up primarily using this network for
final-output of high resolution images or some such.. or test renders at
high res that will be completed relatively quickly
In fact don't even link back to a website, but just have a little status
window available that shows the information, and perhaps a link back to that
users home page and email address if they choose to provide it :)
The user can optionally provide access to his 'rendering' directory so that
others can view the completed images, or past renderings. In this way you're
relatively detached from the web and there is no central repository as such
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
"Peter J. Holzer" wrote:
> I have downloaded the program, but I couldn't find much information on
> how it works, and - most importantly - no source code on the site.
You're right. I will put an access to the source code throught CVS this
week.
You'll see that it's quit simple to get in the code.
Keep in mind that it is still under development.
Gilles Equipe Architectures Paralleles Tel 33 01-69-15-42-25
FEDAK LRI bat 490 Fax 33 01-69-15-65-86
------ Universite Paris-Sud email: fed### [at] lrifr
XtremWeb F-91405 ORSAY Cedex http://www.lri.fr/~fedak
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
"Peter J. Holzer" wrote
> RLIMIT_CPU
> (limits the total CPU usage for one job)
>
> RLIMIT_DATA, RLIMIT_STACK, RLIMIT_CORE
> (limits the memory usage of a single process not all unixes have all
> of these)
>
> RLIMIT_NPROC
> (limits the number of processes per user)
>
> RLIMIT_NOFILE
> (limits the number of open files per process)
>
> Together with quotas, a chrooted environment and a positive nice factor,
> the povray process(es) would be restricted from disturbing their host
> computer.
An other approach we are concidering is to confine the execution in a
sandbox.
Last week we have played with user-mode-linux.
(http://www.user-mode-linux.sourceforge.net)
Idea is to boot a virtual os only to run the application (povray). So
even if Povray is compromised, the user
is still in the sandbox and don't have access to the hosting machine.
Some preliminary test show that this solution
is quite easy to use. More, it has cool features such as defining the
amount of memory (physical and swap)
the process may use, the amount of disk and so on... So it is a very
appealing solution.
But we are still checking security and performance issues.
We are also comparing it to other solution also PTRACE oriented like
Janus or subterfugue (http://subterfugue.sourceforge.net)
> You can leave this to the slave servers: They can check whether there is
> some output produces within a given time. If there isn't, they'll kill
> the process and report failure.
>
Yes, it's the only way, unless that the server could be constantly
overloaded.
--
Gilles Equipe Architectures Paralleles Tel 33 01-69-15-42-25
FEDAK LRI bat 490 Fax 33 01-69-15-65-86
------ Universite Paris-Sud email: fed### [at] lrifr
XtremWeb F-91405 ORSAY Cedex http://www.lri.fr/~fedak
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
|
|
