In the beginning...

The modern era of visualization was initiated by the landmark NSF report, Visualization in Scientific Computing (McCormack, de Fanti and Brown, 1987). This argued that investment in high-performance computing needed a similar investment in visualization: there would be no benefit from the increased simulation power if the results could not be interpreted quickly enough. Visualization was the key.

Reference Model ...

This motivated the development of a number of dataflow visualization environments in the late 1980s and early 1990s. These followed a common reference model (Haber and McNabb, 1990), in which data was progressively transformed in a pipeline of elementary operations - first, into an abstract geometric representation - a visualization - and then from geometry into image, by rendering. 

   Visualization Pipeline

Visualization Systems ...

The first systems to be developed were AVS (Upson et al, 1989)(below left) and apE (Dyer, 1990), but they were rapidly followed by several others: Khoros (Rasure and Wallace, 1991); IBM Data Explorer (Abram and Treinish,1995)(below right); and IRIS Explorer (Foulser, 1995). With the exception of apE, all are still active products, having evolved to embrace new technology and algorithmic developments over the past decade. IBM Data Explorer is now an open source product, IBM Open Visualization Data Explorer; IRIS Explorer , initially developed by Silicon Graphics, is now developed and distributed by NAG Ltd. Interesting new systems have also emerged over the years, for example, the VISSION system. (Telea and van Wijk, 1999) 

An example: IRIS Explorer ...

IRIS Explorer is typical of these systems. A user creates a visualization by selecting modules from a library (shown on left of screen); these are 'wired' together in a pipeline of processes, just as in the reference model. Here we see data being read in (pressure over aircraft wing); an isosurface extracted; and the geometry rendered. 

Extending to Computational Steering ...

A key design feature of these systems is that they are extensible. A user can encapsulate their own code as a module which can appear in the pipeline just like any other module. The parameters of the code can appear as widgets on the module interface panel. This allows a simulation code for example to be encapsulated as a module, and included as part of the pipeline. The reference model easily extends from simple visualization to support the concept of computational steering: simulation control parameters are now the input data at the head of the pipeline. 

   Visualization pipeline with computational steering

Computational steering in practice ... 

An early demonstration of computational steering was the paper by Marshall et al (1990), where a 3d turbulence model of Lake Erie was studied: parameters such as external forces on the Lake were changed as the simulation ran. An extension of the steering concept - to allow backtracking to previous stages in a history tree - was developed in the GRASPARC project (Brodlie et al, 1993). However the potential of computational steering has not yet been fully realised because applications need to be distributed: typically simulations need to run on large compute servers, while visualization needs to run on the desktop. Mechanisms to run individual modules remotely do exist in many visualization systems, but are rather primitive and insecure. It is an aim of this demonstrator project to show that a simulation can run securely on a Grid resource, with visualization on the desktop.

Introducing collaboration into visualization ...

Much scientific research today is done in teams, often geographically separated. This is especially true of large, Grand Challenge projects. The paper by Wood, Wright and Brodlie (1997) shows how dataflow visualization systems can be extended to support multi-site working: each researcher runs their own visualization pipeline but can program an exchange of data and control parameters between pipelines. Thus for example, the geometry from one pipeline can be transmitted to another pipeline where it may be rendered.

Collaborative visualization systems ...

The concept has been realised by Wood in terms of IRIS Explorer, and it is now an integral part of the system. Individual sessions of IRIS Explorer connect to a central server, which manages the exchange of data and parameters between participants. In addition, users can exchange pieces of pipeline - allowing for example a new collaborator to join in easily. Also, pipelines can be hidden by grouping modules into an application with a single interface panel, this panel being shared by all collaborators. Similar ideas have been studied for AVS in the EU MANICORAL project, and at SDSC in the cAVS work. 

      Collaboration using Iris Explorer

Collaborative visualization meets computational steering ...

It is a small step to combine collaborative visualization and computational steering. We take the computational steering model from above, and add in the ability to exchange data and parameters across the Internet. This can be achieved, today, using IRIS Explorer with the COVISA toolkit. 

... and finally the Grid

The final step in the jigsaw (at this stage) is to allow the simulation to execute, not on the desktop, but remotely on a Grid resource. To achieve this, we can split the simulation into two parts: the interface and the engine. The interface runs as a module in the visualization system on the desktop, and communicates with the engine which runs remotely as a Grid resource. 

This demonstrator project has been funded by the EPSRC / DTI e-Science initiative. It consists of a series of individual demonstrators which you can use to help learn how collaborative visualization and computational steering provide important tools for e-Science. 

References:

Abram, G and Treinish, L. An Extended Data-Flow Architecture for Data Analysis and Visualization, Computer Graphics, Vol 29, Number 2, pp17-21, 1995.

Brodlie, KW et al. GRASPARC - A Problem-Solving Environment Integrating Computation and Visualization. Proceedings of IEEE Visualization 1993, pp102-109, IEEE Computer Society Press, 1993.

Dyer, DS. A Dataflow Toolkit for Visualization. IEEE Computer Graphics and Applications, 1990.
Foulser, D. IRIS Explorer: A Framework for Investigation. Computer Graphics, Vol 29, Number 2, pp13-16, 1995.

McCormick, BH, de Fanti, T and Brown, MD Visualization in Scientific Computing, Computer Graphics, Vol 21, No 6, 1987.

Rasure, J and Wallace, C. An Integrated Dataflow Visual Language and Software Development Environment. Journal of Visual Languages and Computing, Vol 2, pp 217-246, 1991.

Telea, AC and van Wijk, JJ. VISSION: An Object Oriented Dataflow System for Simulation and Visualization. Proceedings IEEE VisSym '99, Springer, 1999.

Wood,JD, Wright, H and Brodlie, KW. Collaborative Visualization. Proceedings of IEEE Visualization 1997, pp253-260, IEEE Computer Society Press, 1997.