Hierarchical graphics databases in sort-first

Abstract

Sort-first architectures for parallel rendering use the natural frame-to-frame coherence of primitives on the screen to increase performance by grouping primitives that are close together on the screen onto the same processor. This technique reduces processor-to-processor communications overhead. As primitives move on the screen, they may also be moved to other processors. This migration poses many problems for the editing and traversal of hierarchical graphics databases (HGD's) such as PHIGS. In this paper we describe the major problems associated with implementing an HGD on a sort-first system. We discuss the basics of HGD's and how they extend to parallel graphics systems. We show why bookkeeping for HGD's is a more complex issue for sort-first than for other parallel architectures. We then examine two possible solution branches for this issue and delve into the design choices and implementation issues that arise with either method. The methods differ in the amount of bookkeeping that they do versus the amount of primitive cull-testing that must be performed.