PERFORMANCE
Cross-platform running times markedly differ. For 2,000
HTML questionnaires
of 120 kB each, containing 120 lines (with IDSS definition of a
'line'), 36
variables and a total of 192 characters to be parsed, on a laptop built
in 2003, equipped with an Intel Mobile
Celeron processor running at 2,4 GHz and 512 MB of DDR, the
following results were obtained on generating a separated database
(option -B) :
GNU/Linux | Windows XP | Cygwin (with Windows XP) |
< 20 seconds of processing time (minimum 14 seconds out of 20 tests) 20-25 seconds of real time |
75-85 seconds of real time | > 4 minutes of real time |
Silent mode -W was used to cut down running times in all tests.
Under GNU/Linux, silent mode running times (options -ir -B -e -s -W) are slightly higher than gzip
compression of the questionnaire directory (command 'tar -cf9z'),
yet three times lower than copying the same directory from one
file system to another ('cp -R') and four times lower than bzip2 compression (command 'tar -cf9j'). These are indicative results, which are likely to depend on hardware configuration.
Implementing an algorithm written in
C limited to the core requirements of survey management made it
possible to enhance input parsing performance. Many professional
applications take a different approach, implementing general-purpose
algorithms (notably regular expressions), whose complex routines allow
for more layout flexibility at the expense of higher running times.
Commonly-used interpreted languages
boast high-quality libraries for parsing HTML documents (VBA under
Windows, Python and
Perl in Unix systems) and perform varied tasks; however, IDSS runs faster by an order of magnitude
at least.
Considering these comparative tests, it is advised to run IDSS under GNU/Linux (or more generally a POSIX standard) to process a large number of questionnaires.