[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Please fill me in....

    Date: Wed, 3 Jan 90 13:42:20 CST
    From: Arthur D. Flatau <flatau@CLI.COM>

       Date: Wed, 3 Jan 90 11:00 PST
       Internet-Address:   SPOCK%samson.cadr.dialnet.symbolics.com@riverside.scrc.symbolics.com
       Snail-Mail-Address: American Microsystems Inc. (A wholly owned subsidiary of Gould Inc.)
			   CAD Research Lab.
			   P.O. Box 967
			   Twain Harte, CA  95383
       Phone-Number:       (209)586-7422

	   Date: Sat, 23 Dec 89 17:11:23 CST
	   From: Arthur D. Flatau <flatau@CLI.COM>


	      No.  In order to run Lisp effectively, you need a decent amount of
	      memory (16Mb, I guess) and a local paging disk.  This would bring the
	      total cost of the workstation up to $15-20K.


	   I have a Sun 3/60 with 20Mb of memory.  The main lisp program I run on
	   it is the Boyer Moore theorem prover.  It has some (maybe lots) of
	   declarations.  It runs faster on my Sun (under Austin Kyoto Common
	   Lisp [AKCL]) then on the Symbolics.

       What kind of Symbolics are you comparing to and how much disk/memory?

    To a 3645 with 2 190Mb disks (I don't know which ones they are) and
    8Mb (2 Mwords) memory.

	 I do not have a local paging
	   disk.  However if you run AKCL (or Lucid for that matter) with a large
	   enough Lisp image you lose big during garbage collection.  This is
	   because of the large number of page faults generated.  Lucid seems to
	   be a bit better about garbage collecting, but the same problem will
	   arise at some point.

       Ahh yes, page faults would be deadly for you which is probably why you
       have(need?) 20mb of memory.  But I experienced the same slowdown in my
       sun setup which had a local paging disk.  So you're saying that there is
       a significant difference between the number of page faults on a Sun vs.
       Symoblics during garbage collection?  Or are there more things involved
       in GC (tag bits?) that create problems for "standard" architectures.?

    I normally run the Symbolics with the ephemeral GC on, and the
    incremental GC off.  The difference in paging is (I believe) that the
    Symbolics ephemeral GC mostly looks at pages that you have used
    recently, while the AKCL GC looks through the entire memory.



Yes, of course.  However, I was using Lucid which has some sort of
ephemeral GC.  I don't know how to compare their ephemeral with
symbolics though.  Oh well... Thanks again.