distributed rendering

Christoph Fabianek fabianek at cs.utk.edu
Sun Sep 21 20:40:56 CEST 2003


> > That is definitly much, but I can't do much about this. What happens is
> > that nspovray builds a .ini file which contains all options and a tar
> > file that includes all necessary files (*.pov, *.inc ...). These two
> > files are sent to each server and when it is finished the TGA is sent
> > pack. I don't know about the internal protocol of NetSolve.
>
> No, I was not complaining about the traffic, just some background
> information that nspovray actually transfers data but doesn't show an
> image.

nspovray collects the data until a complete row of tiles is sent back from the 
servers, at this time you see network traffic but no output. When the data is 
here the tiles are composed and written to stdout at once - hope this answers 
the question?


> > If you don't get any output like
> > Downloading output from anakin04.sinrg.cs.utk.edu
> > this is probably because of a firewall. You can either try to open the
> > ports mentioned above or use the SSH transport. To test this create a
> > file e.g. "ssh_hosts" with the following line
> > localhost user /usr/bin/povray /usr/share/povray/include/
> > or something like that. This should also be much faster than using the
> > NetSolve grid (but you need to have a login at the machines and
> > auto-login via ssh)
> >
> > Btw, you may also try nspovray at the command line - let me know what
> > output you get there:
> > $ export NETSOLVE_AGENT=netsolve.cs.utk.edu
> > $ ./nspovray +iicl.pov
>
> No difference.
> Hmm, that's a problem. I have a standard SuSE firewall with no special
> rules, that protects the computer from the outside only. Outgoing ports are
> all open AFAIK. And that kind of firewall configuration is quite common.

Ok, here how NetSolve works if I remember correctly.
- nspovray submits a problem request (povray) to the agent (port 9060) and
  gets back a list of available servers for this problem
- nspovray submits the problem to one of the servers
- the server processes the job and sends back the results <- here it needs the
  ports to be open to accept conncections from outside (I think this means
  "server ports")

Is the SSH transport working?


> > - I like to have options of a logical entity at a single place - even if
> > that means to enter them for each scene (the only option that make sense
> > to move outside would be "nspovray Path" - and you have the file selector
> > to do this quickly. So, no big deal in my opinion.)
>
> It doesn't only mean to enter it for each scene but for each render mode
> you want to use nspovray. These kind of settings have to be permanent.
> Adding a new configure page is no big deal either.

So I will add a page ("Distributed rendering") in the Settings Dialog with the 
"nspovray Path" field - ok?


> > > The feature freeze has been delayed.
> >
> > What are the consequences for kpovmodeler - do you also plan to shift
> > back the feature freeze? Especially, will the distributed rendering go
> > into KDE 3.2?
>
> If my points and questions are solved, then I will add it to KPovModeler
> 1.1. At the moment there are too many open points and problems.

I will keep working... the deadline is still tomorrow?


Greetings,
   Christoph




More information about the kpovmodeler-devel mailing list