Project

Back to overview

Distributed Compositing and Data Management for Scalable Parallel Rendering

English title Distributed Compositing and Data Management for Scalable Parallel Rendering
Applicant Pajarola Renato
Number 129525
Funding scheme Project funding (Div. I-III)
Research institution Institut für Informatik Universität Zürich
Institution of higher education University of Zurich - ZH
Main discipline Information Technology
Start/End 01.04.2010 - 30.06.2011
Approved amount 55'911.00
Show all

Keywords (4)

Distributed Rendering; Scientific Visualization; High-Performance Parallel Rendering; Large-area Display Walls

Lay Summary (English)

Lead
Lay summary
The continuing improvements in hardware integration lead to ever faster CPUs and GPUs, as well as higher resolution sensor and display devices. Moreover, increased hardware parallelism is applied in form of multi-core CPU workstations, clusters, massive parallel super computers and multi-sensor scanner array systems. Consequently this leads also to a rapid growth in complexity of data sets from numerical simulations, high-resolution 3D scanning systems or bio-medical imaging, which causes interactive exploration and visualization of such large data sets to become a serious challenge. Effective and timely analysis of such vast amounts of data has become a major undertaking, and interactive exploration and visualization of huge data volumes becomes one of the most important tools for gaining insight into the structure of the data. However, the emerging large scale data sets can no longer be efficiently displayed with standard visualization systems, which are typically based on high-performance visualization workstations, as the data size greatly exceeds the available graphics capacity. It is thus crucial for a visualization infrastructure to take advantage of hardware accelerated scalable parallel rendering.A major challenge with all high-performance parallel computing frameworks is the optimization of the overall throughput and performance of the system. In parallel rendering, as the raw rendering cost goes down with the number of nodes, the system bottleneck shifts to the synchronization and final image compositing stage. Hence a saturation of the distributed compositing eventually bounds the achievable frame rate despite increasing the number of concurrent graphics nodes. In this project extension we therefore we want to focus on identifying the limiting cost factors and improve overall performance by reducing the cost of the image compositing stage.The second area of extended work includes distributed data management. A parallel rendering application must make sure that the data required to render is loaded on the respective graphics node that needs it for this particular frame. Since this cannot be solved in a generic way for unknown 3D geometry and texture data formats and internal representations we want to focus on a specific form of hierarchical data representation, scene graph approaches such as OpenSceneGraph (OSG) or OpenSG. A major part of this project is thus the realization of an integrated scene graph model and Equalizer integration.
Direct link to Lay Summary Last update: 21.02.2013

Responsible applicant and co-applicants

Employees

Name Institute

Associated projects

Number Title Start Funding scheme
116329 VISS: Visualization Infrastructures for Scalable Scientific Computing 01.04.2007 Project funding (Div. I-III)
121072 High-Performance Visualization 01.04.2008 International short research visits

Abstract

MotivationThe continuing improvements in hardware integration lead to ever faster CPUs and GPUs,1 as well as higher resolution sensor and display devices. Moreover, increased hardware parallelism is applied in form of multi-core CPU work- stations, clusters, massive parallel super computers and multi-sensor scanner array systems. Consequently this leads also to a rapid growth in complexity of data sets from numerical simulations, high-resolution 3D scanning systems or bio-medical imaging, which causes interactive exploration and visualization of such large data sets to become a serious challenge. Effective and timely analysis of such vast amounts of data has become a major undertaking, and interactive exploration and visualization of huge data volumes becomes one of the most important tools for gaining insight into the structure of the data. However, the emerging large scale data sets can no longer be ef?ciently displayed with standard visualization systems, which are typically based on high-performance visualization workstations, as the data size greatly exceeds the available graphics capacity. It is thus crucial for a visualization infrastructure to take advantage of hardware accelerated scalable parallel rendering.Application domains involving interactive visualization of very large data include, among others, volume rendering in bio-medical or oil & gas geological imaging, as well as scienti?c visualization in computational chemistry, astro- physics, and computational ?uid dynamics. Additional examples include data centers for large data analysis and remote visualization, and immersive visualization systems driving display walls, CAVEs2 and auto-stereoscopic displays for simulation, training, and command & control VR applications.ScopeThis proposal is a request for a one year extension of our currently active SNF project 200021-116329 VISS: Visualiza- tion Infrastructure for Scalable Scienti?c Computing (VISS). In VISS we have focused on the development of a new scalable parallel rendering framework called Equalizer that is aimed primarily at cluster-parallel rendering, but works as well in a shared-memory graphics super-computer system. Equalizer improves upon prior solutions in terms of ?exi- bility and scalability by providing a minimally invasive programming model and an abstraction of the graphics layer for synchronized, distributed and parallel real-time rendering.A major challenge with all high-performance parallel computing frameworks is the optimization of the overall throughput and performance of the system. In parallel rendering, as the raw rendering cost goes down with the number of nodes, the system bottleneck shifts to the synchronization and ?nal image compositing stage. Hence a saturation of the distributed compositing eventually bounds the achievable frame rate despite increasing the number of concurrent graphics nodes. In this project extension we therefore we want to focus on identifying the limiting cost factors and improve overall performance by reducing the cost of the image compositing stage.The second area of extended work includes distributed data management. A parallel rendering application must make sure that the data required to render is loaded on the respective graphics node that needs it for this particular frame. Since this cannot be solved in a generic way for unknown 3D geometry and texture data formats and internal representations we want to on a speci?c form of hierarchical data representation, scene graph approaches such as OpenSceneGraph (OSG) or OpenSG. A major part of this project is thus the realization of an integrated scene graph model and Equalizer integration.This project extension is primarily designed to support the completion of the ongoing doctoral dissertation of Maxim Makhinya, who has been involved with the original project from the beginning. A one-year project extension would allow him to complete his dissertation by addressing the above research challenges.
-