Path: utzoo!mnetor!uunet!seismo!sundc!pitstop!sun!decwrl!labrea!agate!pasteur!ames!hc!beta!dzzr From: dzzr@beta.UUCP (Douglas J Roberts) Newsgroups: comp.lang.lisp Subject: Re: Troubles Message-ID: <15762@beta.UUCP> Date: 18 Feb 88 23:41:53 GMT References: <6908@sol.ARPA> Organization: Los Alamos Natl Lab, Los Alamos, N.M. Lines: 59 Summary: Symbolics vs. Suns In article <6908@sol.ARPA>, miller@ACORN.CS.ROCHESTER.EDU (Brad Miller) writes: The discussion was about the processing speeds of a Symbolics 3650 vs a 3670. > Slower? I thought it was about 1.4x *faster*. Our benches show the 3650 to be ~20% slower than a 3570. > And a 16MB 3620 (the same > processor, but slower disks, diffo cabinet) would only set you back about > $55k. *NOT* diskless. Our benches showed that the Symbolics 3620 suffered severe performance (speed) degradation as garbage collection progressed, because of its slow Winchester drives. Initially, the 3620 ran approximately as fast as a 3600, but as GC and paging demand increased performance rapidly dropped off to the point that a benchmark run took 45% longer on the 3620 than on the 3600. Thus the reason for comparison between a Symbolics 3650 and a Sun 3/260: equivalent run times. > If you are going to compare to a SUN, then add disks > to the machine. Or count some fraction of the cost of your sever and > remember that you are paging over the net. Ok. Four user Sun system: Sun 3/280 Server w/2 280 MB drives ~74,600 4 - 3/260 diskless, 16MB ~143,200 Total 217,800 (Ignore misc cables, etc.) Cost / user: 54,450 Four user Symbolics 3650 configuration (believe me, the 3620 is not sufficient for large KEE applications). Cost / user: 75,000 So far, paging over the net with our Suns hasn't hurt us. We don't have any numbers yet, but it's beginning to appear that 1 server can handle 7 - 8 diskless clients running large LISP images before it will start to bog down. > Lets face it, you can't make a lispm out of a box that is running > UNIX. Have you *seen* SPE yet? (I kind of doubt it. We're beta testers, so there can't be too many people outside of Sun familiar with it.) It's pretty damn close to a LISPM. ------------------------------------------------------------------------ Doug Roberts Los Alamos National Laboratory dzzr@lanl.gov