R. Samanta, T. Funkhouser, K. Li, and J. P. Singh. Hybrid Sort-First and Sort-Last Parallel Rendering with a Cluster of PCs. In Proceedings of the Eurographics workshop on Graphics hardware, pages 97--108, 2000.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....handle models with tens of millions of polygons at interactive frame rates. On the other hand, MMR required up to weeks of preprocessing time and expensive high end graphics workstations. Our approach requires only minutes of preprocessing, and works on a cluster of commodity PCs. Samanta et al. [28, 29] have developed a sort first rendering system using a network of PCs. The main focus of their work was on load balancing the geometry processing and rasterization work done on each of the PCs, while we focus on handling very large models. 3 Acquiring the Geometry To acquire the geometry of an ....

.... every 6 12 months; we can easily add or remove machines from the cluster, mix machines of different kinds, and user the cluster for tasks other than rendering; and the aggregate computing, storage, and bandwidth capacity of a PC cluster grows linearly with the number of machines in the cluster [28]. Parallel rendering strategies fall within three main categories, depending on which stage of the rendering pipeline sorting for visible surface determination takes place [21] These categories are sort first, sort middle, and sort last. Sort first approaches divide the 2D screen into disjoint ....

R. Samanta, T. Funkhouser, K. Li, and J. P. Singh. Hybrid sort-first and sort-last parallel rendering with a cluster of PCs. In 2000.

....of PCs. A sorting classification of different approaches for parallel rendering has been described in [MCEF94] Some of the recent work on rendering large geometric datasets has focused on using PC clusters. These include techniques to assign different parts of the screen to different PCs [SFLS00] as well as distributed algorithms for scalable displays [HBEH00] Other cluster based approaches include WireGL, which allows a single serial application to drive a tiled display over a network [HEB 01] as well parallel rendering with k way replication [SFL01] The performance of these ....

R. Samanta, T. Funkhouser, K. Li, and J. P. Singh. Hybrid sort-first and sort-last parallel rendering with a cluster of PCs. Eurographics/SIGGRAPH workshop on Graphics Hardware, pages 99--108, 2000.

....A number of parallel approaches based on multiple graphics pipelines have been proposed. These can provide scalable rendering on shared memory systems or clusters of PCs. These approaches can by classified mainly as either object parallel, screen spaceparallel, or frame parallel [HEB 01, SFLS00] Specific examples include distributing primitives to different pipelines by the screen region into which they fall (screen space parallel) or rendering only every Nth frame on each pipeline (frame parallel) Another parallel approach to large model rendering that shows promise is interactive ....

....depth complexity. For low depth complexity scenes there is little or no speed up, but there is no loss in frame rate as the occlusion culling is performed using a separate pipeline. However, our parallel algorithm introduces a frame of latency. Note also that other parallel approaches [HEB 01, SFLS00] are fundamentally orthogonal to our approach, and could potentially be used in conjunction with our architecture as black box replacements for the OR and RVG rendering pipelines. 7.3 Load Times One of the considerations in developing a walkthrough system to render gigabyte datasets is the time ....

R. Samanta, T. Funkhouser, K. Li, and J. P. Singh. Hybrid sort-first and sort-last parallel rendering with a cluster of pcs. Eurographics /SIGGRAPH workshop on Graphics Hardware, pages 99--108, 2000.

.... tasks among nodes in a cluster, eventually redistributing the resulting non overlapping pixel tiles to drive a tiled display [29, 31] They then extended this technique to allow for tile overlap, creating a hybrid sort first and sort last algorithm that could effectively drive a single display [30]. All of these algorithms required the full replication of the scene database on each node in the cluster, so further work was done to only require partial replication, trading off memory usage for efficiency [28] Although these papers present an excellent study of differing data management ....

Rudrajit Samanta, Thomas Funkhouser, Kai Li, and Jaswinder Pal Singh. Hybrid sort-first and sort-last parallel rendering with a cluster of PCs. Proceedings of pages 97--108, August 2000.

....be sent to which server. Each server renders only those objects that fall into the tile of its associated projector. The latter means that the workload can be unbalanced if the scene is not spread equally over the screen. This could be solved by applying a hybrid sort first sort last algorithm [12]. This would, however, increase bandwidth usage and would complicate matters. Therefore, our initial focus is on the simple sort first approach. In the future, different approaches will be investigated. 3.4.1 The Multiple Copies and Broadcast approaches The two Aura renderers use different ....

R. Samanta, T. Funkhouser, K. Li, and J. P. Singh. Hybrid sort-first and sort-last parallel rendering with a cluster of PCs. In Proceedings of the

No context found.

Rudrajit Samanta, Thomas Funkhouser, Kai Li, and Jaswinder Pal Singh. Hybrid sort-first and sort-last parallel rendering with a cluster of pcs. In Eurographics/SIGGRAPH workshop on Graphics hardware, pages 99--108. ACM Press, August 2000.

....work due to parallelization, scale as more PCs are added to the system, and work efficiently within the constraints of commodity components. Recent work in parallel rendering with PC clusters has focused on strategies that assign rendering work for different parts of the screen to different PCs [18, 24, 25]. It has been shown that communication overheads can be reduced by partitioning the workload in a view dependent manner so that pixels rendered by one PC can be sent to the display directly with little or no depth compositing. However, current methods require either replicating the entire 3D scene ....

....shown that communication overheads can be reduced by partitioning the workload in a view dependent manner so that pixels rendered by one PC can be sent to the display directly with little or no depth compositing. However, current methods require either replicating the entire 3D scene on every PC [24, 25] or dynamically re distributing primitives in real time as the user s viewpoint changes [19] Unfortunately, neither approach is practical for a PC cluster, since the memory of each PC is usually too small to store all the data for a very large model, and the network is too slow to transmit 3D ....

[Article contains additional citation context not shown here]

Rudrajit Samanta, Thomas Funkhouser, Kai Li, and Jaswinder Pal Singh. Hybrid sort-first and sort-last parallel rendering with a cluster of PCs. In Eurographics/SIGGRAPH workshop on Graphics hardware, pages 99--108. ACM Press, August 2000.

....if we have more graphics processors available than there are display devices, this simple, static allocation does not achieve optimal performance. In related work, we have developed dynamic load balancing algorithms for PC clusters, including ones based on sort first [13,12] and hybrid sort last [11] architectures. However, further work is required to develop effective load balancing methods for remote applications. Other data paths: In this paper, we focus on transmission of data from an application to a display. However, other data paths face similar challenges. For example, for ....

Rudrajit Samanta, Thomas Funkhouser, Kai Li, and Jaswinder Pal Singh. Hybrid sortfirst and sort-last parallel rendering with a cluster of pcs. In Eurographics/SIGGRAPH workshop on Graphics hardware, pages 99--108. ACM Press, August 2000.

No context found.

R. Samanta, T. Funkhouser, K. Li, and J. P. Singh. Hybrid Sort-First and Sort-Last Parallel Rendering with a Cluster of PCs. In Proceedings of the Eurographics workshop on Graphics hardware, pages 97--108, 2000.

No context found.

R. Samanta, T. Funkhouser, K. Li, and J. Singh. "Hybrid Sort-First and Sort-Last Parallel Rendering with a Cluster of PCs." Eurographics/SIGGRAPH workshop on Graphics hardware, 99-108, 2000.

No context found.

R. Samanta, T. Funkhouser, K. Li, and J. P. Singh. Hybrid Sort-First and Sort-Last Parallel Rendering with a Cluster of PCs. In Proceedings of the Eurographics workshop on Graphics hardware, pages 97--108, 2000.

No context found.

Rudrajit Samanta, Thomas Funkhouser, Kai Li, and Jaswinder Pal Singh. Hybrid sort-first and sort-last parallel rendering with a cluster of pcs. In Proceedings 2000.

No context found.

R. Samanta, T. Funkhouser, K. Li and J. P. Singh. Hybrid Sort-First and Sort-Last Parallel Rendering with A Cluster of PCs. Proc. of Eurographics Woskshop on Graphics Hardware. August 2000.

No context found.

R. Samanta, T. Funkhouser, K. Li, and J. P. Singh. Hybrid sort-first and sort-last parallel rendering with a cluster of PCs. In 2000 SIGGRAPH/Eurographics Workshop on Graphics Hardware, pages 97--108, 2000.

No context found.

S. Samanta, J. Zheng, T. Funkhouser, K. Li, and J. Singh. Hybrid sort-first and sort-last parallel rendering with a cluster of pcs. In In Eurographics /SIGGRAPH Workshop on Graphics hardware, pages 99--108, 2000.

No context found.

Rudrajit Samanta, Thomas Funkhouser, Kai Li, and Jaswinder Pal Singh. Hybrid Sort-First and SortLast Parallel Rendering with a Cluster of PCs. SIGGRAPH 2000.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC