========================== Frequently Asked Questions ========================== (1) What is the minimum number of processors that I will need on my SGI to run the library? One theoretically. The current distribution is defaulted to 64 virtual processors. (See Makefile on how to change default value or to map virtual to physical processors.) Please be warned that each one of the processes will create a mmap on your /tmp partition. So, if you do not have enough disk space, try not to run too many processes. As you probably know, mmap is not known for good behavior. By being too ambitious you will only end up rebooting your machine. (2) Does the library represent an implicit admission of "flaws" in the single address-space model? No. The library is developed so that we can run iPSC programs on the DASH and to test out codes under a "hybrid" shared- memory/message-passing mode. (3) Is the library inherently inefficient because the development team was restrained by the iPSC interface? Not really. The major inefficiency comes from UNIX and its lack of support for parallel-processing. Which interface we choose has very little to do with it. We will gradually migrate the library out of UNIX-dependent codes into a tailored-made microkernel. (4) Is it trivial to emulate message-passing on a shared- memory architecture? The answer may be 'Yes' on a CS exam. Unfortunately, we are writing a library for industrial-quality codes, and just about anything that can break will break. Memory management, stack overflow, etc... They all need to be taken into account. (5) What is the most popular use of the library right now? Developing iPSC code on an SGI. (6) Can I contribute to the library? Yes. You will lose all of your royalties rights on your contribu- tions, although you can be assured that you will NEVER be short- changed on acknowledgements. In fact, we are making an effort to turn this library into another X-window like consortium. If you can use our codes, why waste your time to write your own? If you develop anything out of our codes, we are glad to integrate it into the public distribution for you. (7) Can my FORTRAN programs use the library? Yes. A FORTRAN-to-C wrapper has been included in the current distribution. See the sample makefile makef77 for linking and loading information. In addition, the FORTRAN library has been in- tegrated successfully with FORTNET, a parallel computing utility developed by the Daresbury Laboratory of the British Science and Engineering Research Council (SERC), which in turns interfaces with a major European collection of parallel numerical algorithms known as PARLANCE. Please contact the Dr. Robert J. Allan at the SERC Daresbury Laboratory directly (rja@cxa.dl.ac.uk) for licens- ing information on FORTNET and PARLANCE.