1 Some comparisons of Algebraic Manipulation on different hardware.
5 Times are elapsed (with cpu times in parenthes for machines where
6 this was substantially different eg suns).
7 Note by elapsed time I wall clock
8 Main culprit was page faults).
9 All times were average of several cycles to get gc information.
10 The large problem gc'd a number of times on the small sun, and
11 so only one computation was performed. All machines were with
12 only one user running at normal priority.
14 Machine. Factor((x+y+z)^10) Factor((x+y+z)^20)
16 sun3/50 27.3 (cpu 9.7) 243.0 (cpu 63.7)
17 too many users to judge: sun3/180 ?16.0(cpu 10.9) ?
18 sun3/260 3.08(cpu 3.08) 19.33(cpu 19.5)
20 Symbolics 3640 5.6 21.3
26 For comparsion the old version using maclisp on dec-20:
27 Quite fast on small problems.
28 Dec20(maclisp) (cpu 5.5) Problem Too Big
32 Notes: The times all are elapsed, and include time for gc. On the
33 lisp machine only ephemeral gc was included. The SUN suffered greatly
34 from page faults: for example the actual run time was 8.05 for the
35 first one, and with a gc every two or three times costing 5 seconds.
36 The rest of the time was spent in page faults, particularly costly
39 Executing on a sun-3/50 with 4meg memory and shoe box. sun operating
40 system 3.2 c compiling of system with debugging on. The common lisp
41 version is by Bill Schelter running in KCL, and running inside gnuemacs.
42 Page faulting was extremely expensive. We have included an appendix for
43 the 3/50 to show the vagaries of the timings due to poor paging.
45 The Maxima in all cases was a full version including lisp compiler
46 and the Maxima to lisp translation facility. Approximately 120 files
47 were loaded. On the very small machine reducing the size of the
48 system, would probably prove beneficial.
50 In all cases the code was the Common Lisp version by Bill Schelter at
51 University of Texas, and distributed by National Energy Software
52 Foundation at Argonne Illinois. KCL is available free of charge by
53 anonymous ftp from University of Texas on rascal.ics.utexas.edu
54 [128.83.144.1] Login as ftp, with no password. See the message
55 kcl.broadcast in the ftp directory for instructions. You must mail in
56 a copy of the form in that message to validate your free license.
57 Both Maxima and KCL include ALL sources.
59 The GCD switch used was SPMOD.
61 Note this is not meant to be a comparison of factoring algorithms, but
62 rather just a simple easily quantified test of complex polynomial
63 manipulation, which has been found to be a good point of comparison
64 when considering different hardware for large algebraic manipulation
67 The Explorer was running release 3.0, and the Symbolics release 7.
68 Note there has been a dramatic improvement in the TI timing with
69 the advent of an efficient ephemeral GC in release 3.0, and new
70 function calling. The time for TI is one half what it was in release 2.0
75 Here is the actual transcript. You can see how the times varied:
76 (C3) expand((x+y+z)^10)$
79 Evaluation took 8.20 seconds (17.93 elapsed)
84 Evaluation took 7.98 seconds (16.98 elapsed)
92 Evaluation took 13.98 seconds (50.25 elapsed)
97 Evaluation took 8.02 seconds (16.57 elapsed)
105 Evaluation took 13.90 seconds (52.20 elapsed)
110 Evaluation took 8.05 seconds (16.65 elapsed)
118 Evaluation took 13.90 seconds (56.88 elapsed)
122 (C16) expand((x+y+z)^20)$
124 Evaluation took 10.10 seconds (13.20 elapsed)
141 Evaluation took 63.77 seconds (243.45 elapsed)