rang%cps45x@CPSWH.CPS.MSU.EDU (Anton Rang) (03/15/88)
Well, I've been following this discussion without comment for a while, but finally decided it was time to add my $0.02. In article <676@kuling.UUCP>, Bo Thide writes: >Well, I just got my FPA installed in my HP9000/350 (multi-user HP-UX) >turning this 4 MIPS machine (4.7 times VAX integer, they say) >into a 3.5 MWhets number cruncher. This is an amazing speed at >an amazingly low total price (less than the equivalent of $50k for the lot). >I am just now installing an Ariel FFT processor (at $1,600) to improve >the speed even further (1024 point complex FFT in 9 ms, 256 point real >FFT in less tan 1 ms). > >Similar workstations are available from Sun, Apollo, Masscomp and others. >This just shows that today's personal "super"-workstations have turned >traditional minis (like the VAXen) obsolete. I agree that "super-workstations" are great for doing arithmetic, both integer & floating-point. There are a lot of reasons to stick with VAXen for some applications, though: 1. DEC's compilers typically do a *much* better job of optimizing than a UNIX compiler on a workstation. Unless you write in assembly, a lot of MIPS can be wasted by poor code. 2. The I/O bandwidth of workstations is often too low for I/O-intensive applications. The UNIX philosophy of "let the user manage his file structure" compounds this problem. 3. VMS provides a rich set of library routines, and an easy means of upgrading without recompilation (shareable libraries). Of course, you can always write your own. 4. (Personal opinion) VMS has a much better user interface, and much better compilers (e.g. extensions/error detection/library handling etc.). There are probably lots of people who disagree with this, of course... One last general note, which doesn't really have to do with this: I heard (at the ACM convention last month) that VMS runs the "typical" program 1.8 times faster than either BSD or Ultrix (on a VAX). I don't know what the "typical" program is, though I suspect it involves a mix of both I/O and computation. <Disclaimer: all opinions & mistakes are my own.> Anton Rang Graduate Student Michigan State University