henry@utzoo.uucp (Henry Spencer) (03/19/88)
> The average program run isn't anywhere near that big, though.
This depends on what you mean by "average". Average of all the programs
used? Average weighted by frequency of use? Average weighted by duration
of run? Average weighted by interactive vs noninteractive? Average weighted
by how much it matters to a particular customer? Some, perhaps most, of the
workstation customers care mostly about how well their big, long-running,
interactive applications perform, and don't care about averages that aren't
weighted to reflect that. All too often the important case is the worst
case, not the average one.
--
Those who do not understand Unix are | Henry Spencer @ U of Toronto Zoology
condemned to reinvent it, poorly. | {allegra,ihnp4,decvax,utai}!utzoo!henrydavidsen@steinmetz.steinmetz.UUCP (William E. Davidsen Jr) (03/21/88)
In article <1988Mar18.174031.653@utzoo.uucp> henry@utzoo.uucp (Henry Spencer) writes: > [...] Some, perhaps most, of the >workstation customers care mostly about how well their big, long-running, >interactive applications perform, and don't care about averages that aren't >weighted to reflect that. All too often the important case is the worst >case, not the average one. While I have no idea how any one site may use their workstations, in looking at the industrial and academic use with which I'm familiar, it sounds as though you are implying that worksataions would be used to do mainframe computing. As nearly as I can determine, workstations are used for graphics, software development, word processing, reading news, and reading mail. They are used to provide a windowing platform with NFS, and generally the programs run are less than 1 min cpu, less than 2MB memory (exculding the graphics display). I realize that some places may not offer mainframe power for large problems, but I doubt that it's common to run really large stuff on a workstation. I'm sure that if others disagree I'll see it here. -- bill davidsen (wedu@ge-crd.arpa) {uunet | philabs | seismo}!steinmetz!crdos1!davidsen "Stupidity, like virtue, is its own reward" -me
gwu@clyde.ATT.COM (George Wu) (03/23/88)
In article <10018@steinmetz.steinmetz.UUCP> davidsen@kbsvax.steinmetz.UUCP (William E. Davidsen Jr) writes: >In article <1988Mar18.174031.653@utzoo.uucp> henry@utzoo.uucp (Henry Spencer) writes: >> [...] Some, perhaps most, of the >>workstation customers care mostly about how well their big, long-running, >>interactive applications perform, and don't care about averages that aren't ^^^^^^^^^^ >>weighted to reflect that. > . . . it sounds as though you are implying that worksataions would be used >to do mainframe computing. > As nearly as I can determine, workstations are used for graphics, >software development, word processing, reading news, and reading mail. >They are used to provide a windowing platform with NFS, and generally >the programs run are less than 1 min cpu, less than 2MB memory >(exculding the graphics display). > > . . . I doubt that it's common to run really large stuff on a workstation. I agree with Henry. But oddly, I mostly agree with Bill, too. Here's my argument/example. Most of my workstation experience has been for using Magic, U. C. Berkeley's VLSI layout system. Jobs tend to be long running editing sessions, ie. interactive. And we (students, this was back when I was working on my BS) used workstations. One slight contradiction here is that Magic was a pretty large program. But for a big high resolution station, we tended to have several megabytes of memory on board anyways. And a few pages usually didn't matter, since they were pretty fast compared to interactive use. I think the key word Henry used was "interactive." When we needed to do really CPU intensive stuff, ie. simulate the circuit, then we'd move it over to something bigger. But most of the stuff was done on a workstation. Overall, both of you seem to be correct. -- George J Wu UUCP: {ihnp4,ulysses,cbosgd,allegra}!clyde!gwu ARPA: gwu%clyde.att.com@rutgers.edu or gwu@faraday.ece.cmu.edu
vanthof@mips.COM (Dave Van't Hof) (03/23/88)
In article <10018@steinmetz.steinmetz.UUCP> davidsen@kbsvax.steinmetz.UUCP (William E. Davidsen Jr) writes: >In article <1988Mar18.174031.653@utzoo.uucp> henry@utzoo.uucp (Henry Spencer) writes: >> [...] Some, perhaps most, of the >>workstation customers care mostly about how well their big, long-running, >>interactive applications perform, and don't care about averages that aren't >>weighted to reflect that. All too often the important case is the worst >>case, not the average one. > [...] > > As nearly as I can determine, workstations are used for graphics, >software development, word processing, reading news, and reading mail. >They are used to provide a windowing platform with NFS, and generally >the programs run are less than 1 min cpu, less than 2MB memory >(exculding the graphics display). Actually, that's what I use the big computers for :-). Seriously, the typical programs I ran on a workstation, generate lots of 20+MB files (some over 100+MB), run for 24+ hours, and would love to have more than 8MB of memory (32 anyone??). Why don't we run these on a mainframe? Cost. As a startup, Mips could not afford to buy mega-$ machine with graphics heads. It was much more cost effective to by several small workstations and work them _hard_. Most of the original CAD verification for the R2000 and R2010 were run on these machines, such as design rule verification, continuity verification, parasitic extraction, logic simulations, etc. On top of that the machines were used for layout and schematic entry. Even our graphics oriented programs are memory hogs, sigh. The real killer part is there are lots of people here who use these machines for large jobs, we're just a bunch of cpu hogs... > > I realize that some places may not offer mainframe power for large >problems, but I doubt that it's common to run really large stuff on a >workstation. I'm sure that if others disagree I'll see it here. As the price drops and the performance increases in the workstations/ super-micro class of machine and as more software gets ported to them :-), you'll see many more _large_ programs gobbling up cycles. What I've heard from my friends on the outside, they are progessively using their workstations for running larger and larger programs. >-- > bill davidsen (wedu@ge-crd.arpa) I realize I may not be a 'typical' workstation user, but this may give an idea of the opposite end of workstation use (abuse? :-)). Dave -- {ames,prls,pyramid,decwrl}!mips!vanthof or vanthof@mips.com (Dave Van't Hof) MIPS Computer Systems, 930 Arques, Sunnyvale, CA 94086, (408) 991-0242 "Never. No, Always question authority." <standard_disclaimer.stuff>
fenwick@garth.UUCP (Stephen Fenwick) (03/25/88)
In article <23698@clyde.ATT.COM> gwu@clyde.UUCP (George Wu) writes: >In article <10018@steinmetz.steinmetz.UUCP> davidsen@kbsvax.steinmetz.UUCP (William E. Davidsen Jr) writes: >>In article <1988Mar18.174031.653@utzoo.uucp> henry@utzoo.uucp (Henry Spencer) writes: >>> [...] Some, perhaps most, of the >>>workstation customers care mostly about how well their big, long-running, >>>interactive applications perform, and don't care about averages that aren't > ^^^^^^^^^^ >>>weighted to reflect that. >> . . . it sounds as though you are implying that worksataions would be used >>to do mainframe computing. >> As nearly as I can determine, workstations are used for graphics, >>software development, word processing, reading news, and reading mail. >>They are used to provide a windowing platform with NFS, and generally >>the programs run are less than 1 min cpu, less than 2MB memory >>(exculding the graphics display). >> . . . I doubt that it's common to run really large stuff on a workstation. > > I agree with Henry. But oddly, I mostly agree with Bill, too. [...] Let me add that I agree with George. Where he and I did our BS work, we used Sun 2's & RT's to run Magic, then 11/780's and Sun 4's to run Spice. So even here, there was the migration of what were traditional jobs for minis moving to workstations. I have also seen an 8600 used to directly drive Tek terminals for chip routing, which is a traditional workstation job. We (Intergraph APD) use Clipper-based boards in AT's to do large simulation jobs (e.g., gate-level simulation of the entire CPU/FPU chip, for >400k simulated clock cycles, or about 1 *week* of cpu time). We used to use an 8600 for this, but: a) we couldn't fit the whole simulation onto the machine without thrashing it; b) the 8600 was much slower even for a smaller model. We have gotten an order of magnitude performance gain by going from a large, mega-buck, multi-user mainframe to a small, cheap (>$15k) Unix-based workstations. So the point is, given the cpu-power now available in a small box for a low price, thinking in terms of workstations only acting as very smart terminals is a bit old-fashioned. However, mainframes can be effectively used where very large amounts of data must be shared in a real-time interactive fashion by multiple users. Steve Fenwick -- \\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\///////////////////////////////////////// My company is not responsible for what I say. I might be... E-Mail route: ...!pyramid!garth!fenwick USPS: Intergraph APD, 2400 Geng Road, Palo Alto, California AT&Tnet: (415) 852-2325 //////////////////////////////////////\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\