Resource-Aware Visualization Using Web Services Ian J. Grimstead, Nick J. Avis and David W. Walker Cardiff School of Computer Science, Cardiff University Roger N. Philp Cardiff School of Physics and Astronomy, Cardiff University Abstract—We present a status report on RAVE, our distributed, collaborative grid enabled visualiza- tion environment. We briefly review our architecture and related work, examine our collaborative support and relate this to an experiment carried out between SuperComputing 2004 (Pittsburgh, PA, USA) and the Cardiff School of Computer Science. Load distribu- tion in RAVE is described and analysed, using a tiled rendering technique to share rendering work- load between machines. Finally, we review various applications that have been extended to use RAVE. I. I NTRODUCTION Increases in network speed and connectivity are promoting the use of remote resources via grid computing, based on service-oriented architec- tures such as the Open Grid Services Architecture (OGSA) [9]. These permit users to remotely ac- cess heterogeneous resources without considering their underlying implementations, both simplifying access for users and promoting the sharing of specialised equipment. With datasets rapidly increasing in size, the vi- sualization of such datasets can quickly overwhelm local computing power. The availability of Grid computing enables users to recruit resources to supply datasets or to assist in their rendering. Grid computing enables co-operation between remote teams at interactive rates and becomes more de- sirable as network technology improves. With this in mind, we present the Resource- Aware Visualization Environment (RAVE), using Grid/Web Services to advertise data sources and recruit rendering assistance from remote resources. II. PREVIOUS WORK Current collaborative visualization systems often make assumptions about the available resources; for instance, COVISE [24] (a collaborative modular visualization package) assumes local rendering sup- port, whilst OpenGL VizServer [18] (collaborative sharing of OpenGL programs) assumes the client has modest rendering capability and instead relies on remote resources. The ARTE environment [14] presents a hybrid approach whereby a full bitmap or geometry may be transmitted, but runs as a single server on a single platform and does not make use of remote resources. glX [23] is a method for rendering data stored remotely over X11, sending unprocessed primitives—requiring both high net- work bandwidth and local rendering. MVEs and Problem Solving Environments (PSEs) are popular tools with visualization, with several projects using this approach. The e-Demand project [5] is implementing a PSE on the Grid, where each module in the environment is rep- resented by an OGSA service, whilst the gViz project [4] is extending IRIS Explorer [22] to be grid-enabled and collaborative (where users can independently control each module of the MVE). For a fuller review of remote visualization appli- cations refer to [3] and [11]. The RAVE system differs from such systems by making best use of available local or remote resources and reacting to changes in these re- sources. In addition, RAVE provides a collaborative environment, a data repository and connects to 3rd party data. III. RAVE ARCHITECTURE We propose a novel and unique visualization system that will respond to available heterogeneous resources, provide a collaborative environment, and be persistent (enabling users to collaborate asyn- chronously with previously recorded sessions). The system must not only react to changes in resources, but also make best use of them by sharing resources between users, and distributing workload amongst resources. The RAVE architecture presented in Fig- ure 1, which we now discuss. An in depth review of the RAVE architecture may be found in our SC2004 paper [11]. A. Data Service The data service imports data from either a static file or a live feed from an external program, either