karl@sugar.uu.net (Karl Lehenbauer) (07/26/88)
> ...My take on this is that > VM may be a lose for the Fortran codes of the world but it is almost > certainly a win for everyone else. In other words, Cray is locking > theirselves into the big iron Fortran club (and out of other places?). Let us not discount that there is a lot of FORTRAN code that has to be run often and quickly and that Cray machines and compilers have been able to do this reliably for years. There aren't many alternatives to FORTRAN for vector machines, even now. The fellow who pointed out that virtual memory is useful for implementing sparse arrays is correct. As is always the case with VM, thrashing must be avoided or performance will suffer dramatically. Unless the virtual address resolution (in the non-page fault case) can be made without extra delays, ever, for the CPU, non-virtual architectures will always be a little bit faster. VM may for this reason be counter to the idea of Speed At All Costs, to which the folks at Cray Research are apparently dedicated. -- -- backups: always in season; never out of style. -- karl@sugar.uu.net aka uunet!sugar!karl