View metadata, citation and similar papers at core.ac.uk brought to you by CORE provided by The University of Utah: J. Willard Marriott Digital Library Guest Editors' Introduction Parallel Rendering Massively parallel computers have emerged as valu­ shared-memory bus architectures. Their work has Chuck Hansen ahle tools for performing scientific and engineering many similarities with cache snooping algorithms. Adavanced Computing computations, far outstripping the capabilities of Paul Mackerras and Brian Corrie describe a independent workstations in hoth sheer floating­ method for exploiting data coherence in image­ point performance and memory capacity. A~ the res­ space volume rendering. They use a eaching scheme Los Alamos National Laboratory olution of simulation models increases, graphics to implement a distributed virtual memory system Los Alamos, NM 87545 algorithms that take advantage of the large memory that effectively minimizes communication by tak­ Internet [email protected] and parallelism of these architectures are becoming ing advantage of data locality during rendering. Although its roots go back many years, the field Tom Crockett increasingly important. This issue of IEEE Parallel Institute for Computer & Distributed Technology highlights some recent of parallel rendering is by no means mature. As you Applications In SCience and work i n parallel computer graphics, specifically par­ will see in this issue, efficient parallel rendering is Engineering allel rendering. an elusive goal, with more questions than answers. Mali Stop 132C It is well known that some computer graphics As parallel computers become more powerful and NASA Langley Research Center problems arc embarrassingly parallel; ray-tracing affordable, the integration of graphics with parallel Hilmpton, VA 23061-0001 algorithms with replicated databases and fractal applications will be a central issue for the visualiza­ Internet [email protected] geometry are classic examples. By replicating data tion community. Efficient parallel rendering on each processor, these problems require little if rithms are a prerequisite to this process. Scott Whitman David Sarnoff Research Center any communication, thus removing a major imped­ CN 5300 im~nt to linear speedup. ACKNOWLEDGMENTS Prince lon, NJ 08543 Generally, however, parallel rendering algo­ The 1993 Parallel Rendering Symposium was held in Internet ,[email protected] rithr1l'i operate on data sets that are distributed conjunction with the Visualization '93 conference, and among the processors. This requires massive com­ was sponsored hy the IEEE Computer Society Tech­ rmll1ication due to the mapping from 3 D "object nical Committee on Computer Graphics in coopera­ space" to 2D "image space." Hence the algorithm tion with ACM SIG Graph. The July 1994 issue of designers must carefully balance computation and IERR Computer Graphics and Applications features five articles four from the symposium - on the graph­ communication. Add to this the nuances of parallel ics aspects of parallel rendering. architectures and a large algorithmic design space, and the task of achieving efficient, scalable perfor­ Chuck Hansen is a technical staff mance becomes very demanding. The problems in member at the Advanced Computing parallel rendering algorithms are similar to those in Laboratory at J .os Alamos l\,"ational Laboratory, and an adjunct faculty all problems on parallel computers: How do we par­ member at the University of New tition a large data set among the memories of the Mexico and New Mexico Institute of computer to minimize communication and network Technology. He received his BS in contention, maximize processor use, and achieve computer science from Memphis State University in 1981 and a PhD good speedup relative to a serial approach? in computer science from the Uni­ To help address these issues, we organized the versity of Utah in 1987. 1993 Parallel Rendering Symposium, which brought together researchers from around the world Thomas W. Crockett is a staff sci­ entist at the Institute for Computer to survey the state of the art in this growing field. Applications in Science and Engi­ The three parallel rendering articles in this issue are neering. He received his BS in math­ updated and revised from papers that appeared at ematics from the Culleg<c of William the symposium. They emphasize a systems view of and lvfary in 1977. parallel rendering, providing into the paral­ lel aspects of computer graphics research without delving into the more esoteric graphics issues. Thomas Crockett and Tobias Orloff look at poly­ gon rendering on a ML\lID distributed memory sys­ Scott Whitman is a memher of the tem, the Intel iPSC/860. They present an asynchro­ technical staff at David Sa rnoff nous algorithm and examine its behavior both Research Center. He received his analytically and experimentally, with an emphasis on PhD and ,vIS in computer science from Ohio State University and his the ttade-offS bClween message size and performance. BS in applied mathematics from Michael Cox and Pat Hanrahan look at the prob­ Carnegie Mellon University. lem of polygon rendering by pixelmcrging. They analyze different pixel merging algorithms based on network bandwidth and develop an algorithm for Summer 1994 7 .