Wilfried Koch
Palanterstr. 46
D-5000 Koeln 41 (Germany)

email: bj030@aix370.rrz.uni-koeln.de


This is a distributed Version of Craig Kolb's Rayshade 4.0

It makes Raytracing very much faster by dividing a frame or animation 
in Squares, which are distributed on a unlimited number of Servers.  


It consists out of two parts.

1. A  run time library (rrlib) basing on Sun's ONC-RPC. It's possible to adapt any render program
(even binaries) to it. 

2. The adaption of Rayshade to rrlib. This adaption works fully transparent to the user. 
			9 rs6000-220 (33MHz)
			1 rs6000-560 (40MHz)	
			1 rs6000-320 (20MHz)
			4 DECStation (20 MHz)
			1 DECSystem  (40 MHz)
			1 Sparc  1 as client  1 DECStation (20 MHz)  Cray Y-MP

Alan Kilian`s aq17.ray		3 hrs 30 min	59 hrs			18 hrs	Squaresize 20

tou.ray(1280x838)		1 hr  45 min		? 		?
Sqauresize 100



Both is tested on IBM RS-6000, DEC-Stations, Sparcs and AIX-370 and should work on HPs and
every unix machine, which has a portmapper, too. The portmapper is neccessary vor NFS.
RRLib does not need NFS by it self.

Next to it there are some examples in the directory exa, the adaption has been
testet with.

Jump into the directory librr and type make.

After everything went well, type cd .. and then ray -h

Take a look into rayshade.srv and exchange the hostname by your
host names. Don't forget to set the correct path for ray.  
Put the client's hostname and the client`s account into
.rhost on the servers account.


RRLib has a lot of functionality and the Rayshade adaption is  very ease to use. Servers
don't need any local file access. You don't have to ftp around your imagefiles, .ray files or colormaps.
All data and results go directly memory to memory over the network.

It's possible to  control all Servers from one machine.
One command is enough to start rayshade and all servers.

RRLib  is load balancing, fault tolerant and needs very less resources. 


If there is any echo refering to bugs or requests for information, how to adapt other render programs
to RRLib, I will make a detailed documentaion how RRLib works and I will clean up the partially 
very nasty code and I will make more statistics, layered and more detailed messages. 

Complex scenes like Alan Kilians aq17 are 8 times faster on 10 machines.
Easier Scenes traced with the -p options are about 6 times faster.
I would be very very grateful, if somebody sends me results about speed gains
with Linda immediatly , i.e. comparisons one vs. more than one worker.  

RRLib is my thesis in CS on 
FH Koeln, Abt. Gummersbach (Germany)
Labor fuer allgemeine Datenverarbeitung
Prof Dr. Horst Stenzel



much fun
	Wilfried
