next up previous
Next: Performance Report Up: Justification for Amount of Previous: Resource Units Per Run

Number of Runs

The total number of runs from all four parts of the proposed studies amounts to of the order of ten, with the exact number determined in part by earlier results: novel initial results might compel us to invest more time in few high resolution runs, or a broader exploration of parameter space. Either way, trial runs on the Origin 2000s at both OSC and NCSA provide us with a dependable estimate of the resources required.

We note the very important point that because of our use of Adaptive Mesh Refinement, low resolution 3D runs can be made on workstation class machines with 128Mb of memory. Such runs have insufficient resolution to address the questions set out above, but provide invaluable insight into which initial conditions and parts of parameter space are most appropriate choices for production runs at OSC. We thus expect to need little or no experimentation, and will make efficient use of OSC facilities.

Runs will require typically 1 - 2 Gb of memory (which is available through a queue - Q95.2g36000 - set up at our request in October 1998). Figure-N shows the results of an exploratory run with a precessing jet; this required 0.7 Gb on the Origin 2000, and employs 2 levels of refinement on a base grid of size tex2html_wrap_inline145 . Although interesting, and indeed capable of giving insight into the dynamics of the flow, the resolution is clearly insufficient. An additional level of refinement alone would lead to untenable memory requirements, but an additional level in conjunction with more selective refinement criteria - restricting refinement to key portions of the flow - would increase the memory required by no more than tex2html_wrap_inline147 .



G. Comer Duncan
Sat. April 3, 1999