HIGH-PERFORMANCE COMPUTING AND NUMERICAL-SIMULATION OF FLAMES

Recently, high performance, relatively inexpensive work stations becam e widely available with scalar peak speeds of 100-600 megaflops (which compares rather favorably with vector supercomputers of not so long a go). While these rates are not seen for problems resulting in sparse m atrices, the rates seen for many problems are quite high. A class of p roblems here is discussed which can now be solved on machines individu als can afford to own rather than just on ones owned by large institut ions. The problem with using a single one of these machines is that th e option of connecting a collection of machines together or buying a p arallel version of the work station becomes irresistable. In fact, dur ing the course of our study we did all of the above. We started on a s ingle machine with a 100 megaflop peak rate (an IBM RISC System/6000 m odel 560 computer). Then we used a farm of the IBM's, an IBM SP1, and finally an SP2. Due to a nice feature of the communications' library w e used, the executables worked on the ethernet or on the fast switches in the SP1/Sp2's without either recompiling or relinking.