rtp1@tank.uchicago.edu (raymond thomas pierrehumbert) (01/05/90)
I had a few moments to spare, and thought I'd set down a few thoughts about what I think Apollo is doing RIGHT, to counterbalance my ample postings about what is wrong with Apollo. My reason for doing this is mainly to provide some encouragement to the people in development at HP/Apollo, lest they give up in despair, or think their work is wholly unappreciated. I think competition has been very good for the workstation industry, and for the workstation user, and I want to see HP/Apollo succeed fiscally, because if they manage to smooth out the wrinkles in their product line (software and hardware), the machines will be unbeatable. So here is what's right: (1) The DN10000 hardware. This is a beautifully designed machine, right from the power supply up through the cpu and bus(es). The decision to used a fast RISC architecture for graphics rather than a dedicated geometry engine was right on target. The vector calls get me speeds of up to 26megaflops on all sorts of useful linear algebra operations (with just one processor, even). (2) Minst. I don't know what installation was like pre 10.2, but with Minst, installation of software is almost as easy as it is on my Mac. I installed Fortran 10.5p, cc, and NFS, and they worked the first time, and it took hardly any time. I ran into some problems with the initial installation of 10.1p on a blank disk, but these were due to a very small (but critical) error in the documentation, which would be easily fixed. In this case, the service people at the 800 number were very prompt and helpful, and managed to sort it out for me (it was a problem relating to starting up glbd, llbd, and rgyd in the wrong order). (3) Compiler reliability. The compilers have a long way to go with regard to optimization, particularly of vector operations, but I have encountered no problems at all with code accuracy (i.e. bad code or wrong results). Given the complexity of what the compiler is trying to do for Prism, this is a little miracle. It is a real contrast with the early Cray compilers, and the CDC Cyber compilers (at all stages of their life). (4) Initial software cost. Apollo has very fair initial prices for their software packages (OS, compilers etc.). The price of the update subscription service is another matter, but you don't have do subscribe if you don't want to. (5) Debugging. DDE is very fine software. It will be even better when and if it runs under X, so I can get at its full glory from any X server. (6) Upgradability and expandability. The DN10000 has the easiest upgrade path, and probably the most open architecture, of any machine in its class. This is the main reason I bought one. It will be nice when the third party developers catch on to this, as the main holdup is lack of drivers for all that hardware you could put in the AT or VME buses. (7) Fortran and C examples. Especially for GPR graphics, this library is a great way to learn how to use the capabilities of the system. I hope 10.2 comes with lots of similar X programming examples (at least in c). I could probably think of more, but that's all I have time for now. Thanks and Happy New Year.