next up previous
Next: Summary of Experimental Results Up: Experiments Previous: Discussion

Comparison with Other Engines

Implementing a new special-purpose Prolog engine, different from the already existing ones, carries a risk: given the level of sophistication of popular Prolog engines, it is useful to check whether the new engine performs comparably with these existing engines, at least for the tasks under consideration here. The efficiency gain obtained through query pack execution should not be offset by a less efficient implementation of the engine itself.

Originally the TILDE and WARMR systems were implemented in MASTERPROLOG. In an attempt to allow them to run on other platforms, parts of these systems were re-implemented into a kind of ``generic'' Prolog from which implementations for specific Prolog engines (SICSTUS, ILPROLOG) can easily be derived (the low level of standardisation of Prolog made this necessary). Given this situation, there are two questions to be answered: (a) does the move from MASTERPROLOG to other Prolog engines influence performance in a negative way; and (b) does the performance loss, if any, reduce the performance improvements due to the use of packs?

TILDE and WARMR have been tuned for fast execution on MASTERPROLOG and ILPROLOG but not for SICSTUS, which makes a comparison with the latter unfair; therefore we just report on the former 2 engines. Table 8 shows some results. These confirm that ILPROLOG is competitive with state-of-the-art Prolog engines.

Table 8: ILPROLOG compared to other engines (times in seconds) for several data sets and lookahead settings.
Bongard-1194 0 7.8 4.74 3.93
Bongard-2986 0 17.8 12.7 9.8
Bongard-6013 0 35 25 18
Bongard-1007 0 0.77 0.74 0.49
Bongard-2473 0 2.07 1.82 1.13
Bongard-4981 0 4.1 3.7 2.2
Bongard-1007 2 7.1 7.5 2.2
Bongard-2473 2 17.7 17.2 4.4
Bongard-4981 2 38 35 8.2

next up previous
Next: Summary of Experimental Results Up: Experiments Previous: Discussion
Hendrik Blockeel 2002-02-26