| Michael Mattias 2004-06-25, 6:50 pm |
| "Robert Wagner" <robert.deletethis@wagner.net> wrote in message
news:40d93f2a.149354897@news.optonline.net...
> This is a re=posting of the sorts demo with one added test: inserting
records
> into an indexed file and reading them back in sequence. Jerry Mouse said
this
> technique is "efficient". In FACT, it ran 15 times slower than a Cobol
sort and
> more than 100 times slower than a fast sort.
One factoid should be presented here: there is not such thing as "a" COBOL
sort. Exactly how the COBOL SORT verb is implemented is compiler-dependent.
Variations in performance across compilers should be expected; performance
comparisons are at valid at best when run in progrms created using the same
compiler; and the different sort algorithms used across the universe of
compilers (or within compilers or external libraries used to effect sorts)
may and will perform differently based on the input data's record size,
record count, key size and key types.
Statements about SORT performance should always carry these disclaimers.
Not to mention, "efficient" is also subjective: speed is but one criteria by
which efficiency may be measured. Memory, disk and CPU usage are others.
The bottom line is, it's easy to find oranges, but it's really hard to find
another apple against which one may compare.
MCM
|