Distributed supercomputing to achieve real-time representation and manipulation of complex scenes

Abstract

Besides the ray tracing technique, the radiosity method is another major approach for global illumination modeling in the field of computer graphics. Since this method needs a huge amount of storage space (both memory and disk) and a long pre-computation cycle, it is not suitable to implement it on conventional workstations. Supercomputers seem to be necessary for such kind of graphics applications. However, some problems appear with the integration of supercomputers and workstations in a distributed computing environment, especially for graphics application. The major problems are: (i) unavailability of standard higher-level mechanisms for building distributed application; (ii) transparent access to supercomputers without remote login or file transfer sessions; (iii) computation balance to make the best use of the performance of supercomputers and workstations; (iv) data transfer/storage strategy between supercomputers and workstations; and (v) conflict between interaction and response time for graphics applications. In this paper we suggest solutions to the above mentioned problems based on the remote procedure call (RPC) and the client/server model, for the radiosity package implementation. After the introduction of the radiosity method, we concentrate on the integration between supercomputers and high-performance workstations. We also convey some general issues related to distributed computing and present the experience with our implementation, such as the application protocol definition, error recovery, code debugging etc. The radiosity package has been implemented at ZGDV in Darmstadt, F.R.G. in a network environment including a Multiflow TRACE 7/300 supercomputer connected to several VAX, SUN and HP workstations.