Xref: utzoo comp.sys.dec:917 comp.arch:7854 Path: utzoo!utgpu!attcan!uunet!lll-winken!ames!decwrl!jumbo!jg From: jg@jumbo.dec.com (Jim Gettys) Newsgroups: comp.sys.dec,comp.arch Subject: Re: DECstation 3100 info. Keywords: 3100, Advantedge 2000 Message-ID: <13508@jumbo.dec.com> Date: 15 Jan 89 22:03:54 GMT References: <979@isieng.UUCP> <85330@sun.uucp> Reply-To: jg@jumbo.UUCP (Jim Gettys) Organization: DEC Cambridge Research Laboratory Lines: 73 I always get amused by Sun folks pushing binary comptibility, when Sun has sold at least (depending upon how you count) three or four or five architectures (680x0, 80386 and SPARC) and has never maintained even complete compatibility between any two series of machines, even within the same family of processors (Sun 1, Sun 2, Sun 3, Sun 4, Sun 386i), and often even between releases of their OS. Maybe they've finally learned the lesson Digital learned over a decade ago... During the same period, Digital has produced 6 additional implementations of the same VAX architecture, fully binary compatible with each other, to go along with the three which precedes Sun's existance. And with very limited exceptions, there has been binary upward compatibility on both VMS and Ultrix during this entire period. Throwing stones in glass houses is usually a mistake. We certainly carefully considered which byte order the PMAX (DECstation 3100) should be run in; in fact, early prototypes have a jumper (never used) which determines the byte order. Now let us examine what various people have suggested; that we should be selling UMIPS on a big endian MIPS machine. I believe those same people would be castigating us about offering two Unix versions, incompatible with each other, of differing byte orders, maximizing the trouble to port to DEC hardware. (Not to even mention the problems this would cause us internally). No matter what we did, it was not possible to make everyone happy. There are certainly good arguments to the byte order question; both sides have cogent arguments on their sides; just look at the wasted network bandwidth in this and other newsgroups on the topic. And being one of the authors of one of the more portable pieces of software in the world, I am completely familiar with the problems of porting software, having spent significant amounts of my time while a DEC employee making sure it would run on RT/PC's, Suns, and other machines. So we went with the byte order which, though us Unix bigots may not like it, is most compatible with the most machines and software in the world, the Intel based PC's (next to which, the hundreds of thousands of VAXes pale, though internal to Digital it is a real issue; for example we have DECnet at FRS, which probably woundn't have made it if the byte order were the other way). I have already recieved mail from one major third party software vendor who was very happy to hear the machine was the same byte order as the PC, as it was going to aid his port from the PC to the DECstation 3100. But as usual, there are arguments on both sides, as another note from a different vendor in this newsgroup suggests. And the choice certainly aids other ports and compatibility with the VAX; for example, the entire core client X distribution compiles and runs on the DECstation 3100 without any ifdef's in any code (you do have to fool it into thinking it is a VAX, as at the moment there are architecture ifdef's for mips in the imake code which really should be UMIPS ifdefs; fixed we hope in X11R4). The port of X to the PMAX was by far the easiest I know of (having participated in ports to the Sun, RT/PC, and a bit to the Apollo and CCI over the last 3 years, though a CCI running 4.3BSD was about as trivial back in V10 days). And I will be happy when OSFix makes real binary compatibility possible between vendors using the same architecture/byte order; I certainly understand the problem of the 3rd party vendors. But at the moment, this was the way we could give them the most REAL compatibility to help them until this day actually arrives; it certainly doesn't exist today for anyone's version of Unix. So, on the issue which is one of the biggest portability sticking points, Digital is at completely consistant; all of our current machines are the same byte order (PC, VAX, MAX). Ultrix on the DECstation 3100 looks essentially identical to that on the VAX, and an amazing amount of code recompiles and runs without change. Jim Gettys Digital Equipment Corporation Cambridge Research Laboratory Cambridge Massachusetts, 02139