Reading 1GB of log files from disc is already slow, independently of how fast Perl is with processing the data - it's still likely IO bound. So the solution must be not to read the whole log file in response to an action from the web interface.

To further the above: Think database. The work process you want is 'Read/compute/store' then 'query/display'. (This assumes you don't want exactly the same statistics every day, in which case just generate a static web page with that when the log files roll over.)

Databases work with massive amounts of data quite well: I deal with multiple logs that are about 1GB compressed every day, and once they are parsed and in the database, data return is nearly instant.

And, luckily enough, Perl has good database support for just about any database you'd ever want to use (and a few you wouldn't).

Maybe you can pre-process the log file then quickly display the results when queried. I have processed website access logs that were about 1GB daily. The reports were detailed and could be accessed in a few seconds, despite it taking several hours to produce them.

If your analysis is not I/O bound (or even if it is) you can prototype your analysis in Perl, then use profiling to identify the critical bits and improve them.

For your current problem, as everyone else said "precompute" is probably the best answer.

You could probably do something complicated to map/reduce the statistics gathering in parallel across sections of the file, but it's probably not worth it.