Shell Programming and Scripting

Yep, you're right. Putting double quotes around the EXCLUDED expression should fix it and since it's a numeric comparison that's probably a better solution that using the [[ ]] as a string comparison. But I suppose they're about the same in this case.

That seems a pretty reasonable size, although the OP didn't say how many records of each he actually has.

While playing with my ksh script I was surprised to find that the associative arrays in ksh don't have the content limit that indexed arrays have, i.e. an indexed array can only have up to 4096 elements but I was able to put 500k elements into an associative array without problems...

I'm guessing that each of these test times is a "second run" test so that the I/O of bringing the executable and libraries into memory has been factored out. If that's the case, I must admit that I'm quite surprised by the awk results. I've not seen awk produce faster runtimes compared to perl when large datasets are processed, yet your numbers show a 2:1 advantage for awk.

I wonder if it would be faster to remove the chomp() in the first perl loop and simply do the lookup by adding a \n in the main loop? I have to assume that concatenating a newline on each record of input would have to be slower than stripping the newline from the exclude list, except I notice that you've got an exclude list 60% larger than the input file so maybe that's a factor?

Very interesting. Thank you for posting those numbers as I now have something to investigate between other projects!

Thank you too Azhrei, that was an interesting discussion.
I had ran some tests as well against a large inputfile BTW.

Code:

Seconds  terminal file  /dev/null
------------------------------------
ksh93    27,15    30,23    25,03
perl     12,89     3,54     3,29
awk       8,15     1,75     1,69

You were right, Perl is much faster with bigger inputs, although not 100 times faster. I thought the differences would be smaller. Well they were small with terminal output, but not with file output. I think the difference between terminal output and output to file is remarkable in the case of Perl. I tried shcomp that compiles ksh93 but that did not make it any faster..

Interesting numbers. Thanks. But you didn't say what your input file sizes were for your tests...?

The terminal output will be constrained by the efficiency of the terminal driver and, assuming you're using a network connection, the efficiency of your network stack. The output to /dev/null is a great test since it eliminates output costs from the processing entirely, although the OP still must perform output to save his data. It gets ugly to test I/O to a filesystem, however, since a fragmented fs will be slower. Ideally, multiple runs would be performed with the first set of results thrown away, and with successive executions overwriting the original data in the file (preventing new data block allocations from being required). Such testing isn't appropriate for the problem at hand, although they might make for some more interesting discussion.

And none of the above factor in usability-related issues, such as the Perl script automatically creating .bak files from the input files.

I think it's time to look at specifics of each execution platform. For example, what version of Korn shell are you using (for example, turn on vi editing mode and type <Esc>^V to see the version number), what version of Perl (just perl -v), and what version of awk (not sure for this one?). Then I'm going to look into this some more when I get some time later in the week.

Interesting thread indeed.

Simply awk --version and while you are at it, give mawk a try if available on your system. mawk interpretor is often much faster than straight awk.

Doesn't work on Mac OS 10.4.11, but strings found version 20040207; probably the BSD version on my machine and you've got the GNU version. I wonder if that's a significant factor? I'd be willing to bet that no one has tweaked at the BSD version of awk since BSD 4.4Lite was released.

I've got Korn shell "M 1993-12-28 p", and my Perl is 5.8.6 with some security patches. But no answer from anyone else on versions.

I used 425412 infile records against 215 excludes (on a laptop Pentium III 850 MHz)

Perl: v5.8.8 built for i486-linux-gnu-thread-multi
Ksh: 93s+20071105-1 The real, AT&T version of the Korn shell
awk: mawk 1.3.3 Nov 1996, Copyright (C) Michael D. Brennan

compiled limits:
max NF 32767
sprintf buffer 1020