(ed.). Gravitational waves (IOP, 2001)(422s).

Cactus computational toolkit 389capability of handling a wide range of dynamical time and length scales.• Challenge in interactive computational science. In spite of the incredibleadvances in computational science, most simulations are still done in a veryold-fashioned way. Jobs are submitted in batch mode, data are output,results are studied, and the process starts over again. This is a very timeconsuming and cumbersome process. In order to really take advantage oflarge-scale simulation as a tool for computational scientists, it is necessaryto develop new techniques that allow one to conveniently make use of theircomputational resources, wherever they may be, to interactively monitor thesimulations with advanced visualization tools, perhaps in conjunction withtheir colleagues in different parts of the world, and to interactively adjust thesimulation based on the observed results.All of these issues lead to important research questions in computationalscience. Here we give an overview of some of our effort in these directions,focusing on performance and coding issues on parallel machines, and on thedevelopment of a community code that incorporates all the mathematical andcomputational techniques described above (and many more), in a collaborativeinfrastructure for numerical relativity.18.6 Cactus computational toolkitThe computational and collaborative needs of numerical relativity are clearlyimmense. To develop a basic 3D code with all the different modules, includingparallel layers, adaptive mesh refinement, elliptic solvers, initial value solvers,gauge conditions, black hole excision modules, analysis tools, wave extraction,hydrodynamics modules, visualization tools, etc, require dozens of person yearsof effort from many different disciplines (in fact, such a feat has still not beendone by the entire community!). Different groups often needlessly repeat eachother’s effort, further slowing the progress of the field. The NSF Black HoleGrand Challenge was a first attempt to address this problem, and an outgrowthof that effort led to the development of the ‘Cactus’ Computational Toolkit(CCTK), developed by the Potsdam group, in collaboration first with NCSAand Washington University, and now with a growing number of internationalcollaborators in various disciplines. Originally designed to solve Einstein’sequations, the CCTK has grown into a general purpose parallel environment forsolving complex PDEs [133–135] that is being picked up by various communitiesin computational science. Here we focus on its application to Einstein’s equations.Cactus is designed to minimize barriers to the community developmentand use of the code, including the complexity associated with both the codeitself and the networked supercomputer environments in which simulations anddata analysis are performed. This complexity is particularly noticeable in largemultidisciplinary simulations such as ours, because of the range of disciplines thatmust contribute to code development (relativity, hydrodynamics, astrophysics,