The change in developed society brought about by the computer revolution shows no sign of abating. The daily utilization of computers and their communication capabilities has altered the way people communicate with each other and organise their lives. As more and more of the world around us become reliant on technology, so we are faced with the challenge of continuing the development of services and ideas whilst managing and adapting existing facilities.
Computer visualization provides hidden perspective and depth to users. Computer graphics technology that was, not long ago, hugely expensive now enables anybody to sit at home and experience a real-time, interactive virtual world of high detail and rea lity. As occurred with computation some years ago, the desire for ever more power has led to parallelisation of graphics processing, enabling ever more extraordinary graphics to be produced in real-time. In line with visualization, the development and spread of networked computing enables the world to work in fundamentally different ways. We are rapidly transiting to a world of immense connectivity in which the challenge is to offer new and existing services in an accessible manner. This is the world of Grid computing.
We have developed a system, called jgViz, that provides an end-to-end user experience in real-time, interactive visua lization in the Grid world. This is an area which is often overlooked by other projects in Grid visualization, which have so far focused on non-interactive scientific visualization. jgViz uses the Grid information services and runtime components of the Globus Toolkit, alongside the Chromium parallel graphics system to enable existing standards-based applications to be used in, and gain from, the Grid environment.
The information model we have designed registers and advertises the presence and capabilities of available jgViz servers through an efficient syntax in a number of Meta-Directory Service (MDS) servers. Our client implementation discovers resources through this hierarchy and communicates with the jgViz servers themselves in order to learn updated statistics on server status in terms of loading and network latency. This information is utilised by the scheduling system of the client to construct a para llel graphics pipeline on available
resources, to suit a user's high-level requirements. The pipeline created is scheduled to provide best performance in either a sort-first tiled-wall or sort-first readback configuration.
To launch a pipeline, we utilize the Grid Resource Allocation Manager (GRAM) protocol for job submission, and the GridFTP and Global Access to Secondary Storage (GASS) protocols for data transfer. Once a pipeline is up-and-running, jgViz will monitor it in order to detect when external factors reduce pipeline performance. When such an event occurs, jgViz will stop the pipeline, reschedule it among the available jgViz resources and restart it.
We have carried out an in-depth study of the performance of the jgViz system, in terms of jgViz-specific capabilities and the limiting factors of such distributed graphics pipelines. We find several things: network performance is the most critical factor with Fast Ethernet not offering acceptable jgViz performance and network latency being key; the 'double-hit' of latency alongside slow readback performance in a readback pipeline restricts the usability of such setups; and that jgViz monitoring does not interfere with the usual operation of a pipeline but does trigger when pipeline performance is impinged in some way. Given
these limitations, we conclude that a general-purpose, real-time Graphics pipeline can be provisioned using Grid technologies.