Date: Wed, 16 Dec 1998 09:56:19 -0600 (CST) From: "M. L. Dodson" <bdodson@beowulf.utmb.edu> To: Benedikt Stockebrand <bs_13943_34262@adimus.de> Cc: hackers@FreeBSD.ORG Subject: Re: Fortran in the base system (was Re: sysinstall) Message-ID: <199812161556.JAA01638@beowulf.utmb.edu> In-Reply-To: <sa7af0ol69i.fsf@adimus.de> References: <199812160645.WAA72686@troutmask.apl.washington.edu> <sa7af0ol69i.fsf@adimus.de>
next in thread | previous in thread | raw e-mail | index | archive | help
Benedikt Stockebrand writes: > I've really tried to stay out of this discussion, but what the fsck... > And I was willing to let the matter drop, but the flame content of the replies to Steve Kargl's comments seem unwarranted IMO. So, let me point out to those who seem so unsympathetic (and I am NOT ANY LONGER SUGGESTING THE ADDITION OF FORTRAN TO THE BASE SYSTEM -- shouting done on purpose!) the basis of my suggestion (note that it was only a suggestion for consideration, done in what I thought was a very polite and helpful way, if you would be so kind as to verify from the archives) was that operating systems which might be considered "friendly" to large scale computation have, as a POLA issue, optimizing Fortran compilers either as part of the base system or as a very carefully integrated addon package, e.g., Irix of various flavors or Solaris. Now I'm going to try to justify why a port does not fall into that category, so if your eyes have begun to glaze over, hit the delete key now. Those of you that are still with me should know that my objective is only to add FBSD to the list of operating systems known to be friendly to number crunchers. The serious computation community is a small, intelligent (but often not well versed in Unix nuances) "market". Hell, they even thought VMS was OK. I just hate for all the press to go to the beowulf guys in that "market" or to have to always justify myself when I tell my fellow members of the "market" my choice is FBSD. And the default f77 may be O.K. to compile that old Adventure game you found in the back of the closet last spring or for computer science students to do their homework for their comparative programming languages course, but it isn't making it for serious scientific work. Now on to the meat of things. The crux of the issue is that people in this "market" frankly don't give a damn about operating systems, my dear. In fact, they look on computers as a necessarily evil even though computers are central to their work. So all of the arguments about the superiority of design, of the development model, etc, are totally lost on them. The one argument that they will listen to goes something like this: "I can tell you how to avoid buying that $25000 Octane and get the same capability on a $4000 PC". "Great, if I don't have to buy the Octane I can use the money for that new probe we've been wanting for the NMR spectrometer! Tell me what to order!" I give them specs on Dell's latest dual processor gadget, tell them how to get a Walnut Creek CD, tell them about XIG, etc. Now comes the crunch. They buy it, then email me to help them set it up. I go over the procedure, then they start to use it to continue development on their NMR data to molecular structure analysis software that they have had in continuous development by graduate students and postdocs for 20 years (none of whom ever had a computer science course). The big guy in the group emails me that the analysis system is dog slow after they got it to compile, even though my benchmarks say that this machine should be equivalent to an R10000 Octane on compute-bound jobs. I ask him if they compiled it with g77, and he emails me back saying "What's that?" I go through all the hooha about the ports system. There is a week or so pause, then he emails me back with something to the effect that they have decided to install Windows on the new Dell and give it to a secretary. It will make a hell of an email reader and Microsoft Word machine, what with the 512MB of memory and all. And, by the way, the next time I have some scatterbrained idea about how to save money, keep it to myself. Now I have two choices: accept the loss of credibility or "go do it for them", causing my own research to suffer along the way. I'm a real scientist with real career goals of my own, not a freelance system administrator for everyone on campus. The point is that these people have other fish to fry. Learning the nuances of a new version of Unix is a big deal to them. They might accept it if they can save money, but any obstacle, any difference from the way they are used to doing things, will cause some of them to drop out. So, what I was really asking for (and I probably phrased things so poorly that people missed the point) was the "carefully integrated" Fortran compilation environment alluded to above. I never meant modularity should be destroyed along the way or any restrictions should be put on the ongoing design process or that innocent bystanders should sacrifice 200MB of disk space to me. Maybe a "good" port is the way to go. I maintain, however, that having two gcc compilers after the installation, even though the correct one can be selected by manipulation of the path, is not productive to the objective I enunciated above. If this argument seems weak, so be it, I guess you had to have been there to appreciate it. But in any case, try to keep the finger pointing and accusations of dark motives to a minimum, O.K.? Bud Dodson -- M. L. Dodson bdodson@scms.utmb.edu 409-772-2178 FAX: 409-772-1790 To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-hackers" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199812161556.JAA01638>