A Parallel Remote Visualization Server for Clusters

Recent research has shown that scalable, high-performance graph-ics systems may be built from commodity graphics hardware run-ning on clusters. The rendered images are then displayed on nodes directly connected to the local system area network. However, many application scenarios require remote display across low speed networks. In this paper we describe a remote, image-based, dis-play system for a central, scalable visualization server. The remote display system involves two components: an image merging stage that assembles tiles from a logical framebuffer distributed across many nodes, and a compression subsystem that compresses the fi-nal image before transmitting it over a lower speed network. In our system the number of rendering and compression nodes may be in-dependently varied to meet the demands of the application and the constraints of the compression and networking technologies. We leverage existing multimedia libraries, in particular, DirectShow, which allows 3D graphics to be integrated into a wide range of In-ternet streaming capabilities, such as video filters and compression algorithms. Using current technology, we demonstrate interactive, low-latency rendering and display of 1024 by 768 images across a 100Mbit network at approximately 13 Hz