lseward@randvax.UUCP (Larry Seward) (11/16/84)
A short report: On the Design and Performance of the Maple System is available from University of Waterloo giving a brief overview of the architecture and comparing Maple, Macsyma and Reduce. Comparisons are given for computing determinants, finding GCD's, solving systems of equations, and for a set of miscellaneous problems. Not suprisingly Maple outperforms Macsyma and Reduce on this self selected suite of problems, in some areas perhaps justifiably so. What is more important are the reasons why the systems vary. For example on the GCD suite REDUCE does poorly because a) the flag for the comparable algorithm (EZGCD) was not turned on, and b) the set of problems used all had trivial (although random) factors for which the default algorithm in REDUCE is known not to perform well. It is the default because it is robust, not because it is efficient. The performance of Maple illustrates the strength of it's multple algorithm architecture. But the problems with the report also illustrate the problems with relying on benchmarks: it is hard to evaluate the benchmarks without understanding the reasons why the systems differ, and then deciding if those reasons are applicable to the problems one intends to work with. For example for many (most?) applications trivial GCD's do not exist, and the GCD suite is not representative. Using test suites to evaluate systems is only valid to the extent that the tests themselves contain documentation about the features they are probing for. The report can be obtained from: Symbolic Computation Group Department of Computer Science University of Waterloo Waterloo, Ontario CANADA N2L 3G1 Larry Seward