Yes, I fixed two significant bugs in your code

That is an outright lie. There simply aren't "two significant bugs" in that code. Neither of those trivia you've identified make the slightest bit of difference to the reproducible data posted.

1) Since I can't reproduce any of the "data" posted in the node with the bugs (since I still don't know what your input file was, other than its size), I wonder what "reproducible data posted" you are talking about.

2) The "trivia" of leaving off /g had a huge impact on performance for me. If I could reproduce your performance data, then I could determine by how much the lack of /g changed it for your secret input file. After making my own input file, the lack of /g made a 10x change in performance (for your code).

3) I was repeatedly quite clear that the two other bugs did not significantly impact performance (which is the data you posted). They were significant in that they caused incorrect output to be produced which was not acceptable to me since I didn't want to do a bunch of benchmarks and then notice that I wasn't actually doing the same work (like you did).

I don't know how you came up with your "2x slower" [....] Since you won't post code

I used your code. The only code I didn't fully post was trivia such as for generating the input and gathering the timings. If you can't figure that out without me spoon-feeding it to you, then here:

time perl -le'print "a"x1024 for 1..100_000' > 1kLines.txt
time perl -pe's/a/A/g' 1kLines.txt > output.cmp
time perl bukQueue.pl 1kLines.txt > output.tmp
diff -sq output.cmp output.tmp
[download]

and real timings he (and anyone; 'cept maybe you) can verify for themselves.

Oh, I didn't realize you had distributed your input file to the OP (and everyone except me). The change in performance is drastically impacted by the input used. For the stupidest-possible benchmark, you can use this instead:

time perl -le'print "z"x1024 for 1..100_000' > 1kLines.txt
[download]

which gives me 10x CPU consumption difference (still nowhere near your original "reproducible data" of "less than 2 minutes" vs "~4 hours").

The more realistic but still silly comparison gave me results similar to:

time perl -le'print "a"x1024 for 1..100_000' > 1kLines.txt
real      0m00.685s
user      0m00.228s
sys       0m00.460s
cpu       0m00.688s

time perl -pe's/a/A/g' 1kLines.txt > output.cmp
real      0m44.254s
user      0m43.323s
sys       0m00.924s
cpu       0m44.247s

rm -f output.tmp
time perl tyeQueue.pl 1kLines.txt
real      0m52.719s
user      1m32.618s
sys       0m02.692s
cpu       1m35.310s
diff -sq output.cmp output.tmp
Files output.cmp and output.tmp are identical

time perl bukQueue.pl 1kLines.txt > output.tmp
real      0m49.997s
user      1m35.598s
sys       0m01.696s
cpu       1m37.294s
diff -sq output.cmp output.tmp
Files output.cmp and output.tmp differ
[download]

CPU consumption goes from just under 45 seconds to a bit over 90 seconds. 2.2x more CPU used.

Feel free to scale up to 3.5GB, change the line contents, and reproduce your "2 minutes vs 4 hours" data but this time in a way that anyone with hours to waste on rather meaningless benchmarking can actually reproduce.

When you do that, also show some benchmarks for your chosen line contents but a small fraction of the 3.5GB so people with less copious time to waste can chip in. And then also explain your guess as to why the smaller file gives 2x..10x CPU consumption difference while your 3.5GB file produces a 120x difference (or whatever it is).

Since you keep emphasizing "120x" as being "my" number, perhaps you could also tell me what number you come up with for "<2 minutes vs ~4 hours". You said "my original figures are based upon actual cpu usage". The first performance figure I see from you is "That will take ~4 hours to process a 3.5 GB file". That sounds like elapsed time to me. So I bet the "120x" is not accurate. It was convenient short-hand and should be in the general ballpark based on "Neither of those trivia you've identified make the slightest bit of difference to the reproducible data posted".

It'd be nice if rather than just hinting that "120x" is an unreasonable interpretation that you would actually provide what the clear performance differences are for your original runs (the "<2 minutes" and "~4 hours" runs). If it was actually "~100x CPU usage" or "~50x elapse time (using two cores)" or whatever, just state it clearly.

And, yes, I haven't posted the code for tyeQueue.pl. As you can see, the performance differences compared to bukQueue.pl are trivial. The output from it is correct, however. I played around with a bunch of things in tyeQueue.pl and they made no difference in performance. So there is no point getting distracted with one particular version of that trivia.

Finally! Some real data to work with.

Just so we can compare like with like, here are your data & tests run on my machine:

<Reveal this spoiler or all spoilers in this node or all in this thread>

So, even with your suspiciously unrealistic choice of dataset, using 1 worker thread requires 15% more cpu & with 4, 26% more. (And that's the original point, but I'll come back to that.)

But replacing every character of every line is ... let's just say extremely unrealistic for now. So let's see what happens when we use what is only a very slightly more realistic dataset with a 50% duty cycle:

<Reveal this spoiler>

Now the differences are rather more obvious. Ranging from 36% more with 1 worker, and 93% more for 4 workers.

And how about the length of those lines? I'd bet that of the data stored in text files around the world the vast majority of it has line lengths far shorter than 1K each. Even really huge datasets with really huge 'records' tend to wrap them at some terminal-friendly limit. Eg DNA & FASTA files.

So let's do something about that also. Cut the line length to a more reasonable 100 bytes and increase the number of lines to 1e6 to maintain approximate parity with the dataset size:

<Reveal this spoiler>

So, now with a dataset somewhat more likely to reflect the norm, we've got 1 worker taking 422% and 4 workers taking 1155% more. That's 4 to 11 times as long. Still a far cry from "my 120x" I hear you cry, but bear with me.

When I came to appraise the possibilities I looked for an existing file of roughly the right size and found a file I'd generated for some similar purpose. 5GB of random 'phrases' in typical 'text file sized' lines. Perfect!:

C:\test>dir phrases.txt
28/12/2010  00:09     5,033,440,760 phrases.txt

C:\test>head phrases.txt
hexagonal monstrance fronded repand trouped modelers
fragged foresighted nescient epilogue athwart venging brickyards
jeweled manometer telium telegony apparitors imperfects deontic impuni
+ty
totemists lire summered nighthawks leucocytes bronchodilators
fissionable collapses luminances masquer notated
giaour overhanded outtraveled moneys dourines undauntedly cockaded sub
+ventions
interrogatives ya prereform tattooing tablemate thunderclouds
numbats preprogram pulque palanquin hobbled
locoing faced overshot nickeled minimum pronunciations pressingly eagl
+ing
lovages fatalities churnings barned treys bombproof refillable
C:\test>perl -pe"s/a/A/g}{warn join' ',$., times" phrases.txt > output
+.cmp

C:\test>perl -pe"s/a/A/g}{warn join' ',$., times" phrases.txt > nul
100000000 159.542 7.878 0 0 at -e line 1, <> line 100000000.

C:\test>junk71 -T=1 phrases.txt > output.cmp
Started Thu Mar 24 04:26:07 2011
1000000 28259.7636234697 at C:\test\junk71.pl line 30, <> line 1000000
+.
2000000 28300.5518302086 at C:\test\junk71.pl line 30, <> line 2000000
+.
3000000 28301.3527797113 at C:\test\junk71.pl line 30, <> line 3000000
+.
4000000 28299.3505386885 at C:\test\junk71.pl line 30, <> line 4000000
+.
5000000 28287.2627896293 at C:\test\junk71.pl line 30, <> line 5000000
+.
Terminating on signal SIGINT(2)

C:\test>junk71 -T=4 phrases.txt > output.cmp
Started Thu Mar 24 04:32:15 2011
1000000 14366.7839914658 at C:\test\junk71.pl line 30, <> line 1000000
+.
2000000 14388.0751686352 at C:\test\junk71.pl line 30, <> line 2000000
+.
3000000 14387.2471354457 at C:\test\junk71.pl line 30, <> line 3000000
+.
4000000 14366.3711996768 at C:\test\junk71.pl line 30, <> line 4000000
+.
5000000 14356.1826309399 at C:\test\junk71.pl line 30, <> line 5000000
+.
Terminating on signal SIGINT(2)
[download]

And so I saw the one-liner complete in (well) under 2 minutes. And from the very consistent lines/sec output, I projected (100e6 / 14387) = 6950 seconds or 115 minutes. If you (care to) remember that my original post contained no timing code, so the timing was done rather more crudely. Hence my factor of two exaggeration.

But, and here is the real point. Never in any of these tests, yours or mine, your dataset or mine, have any of us found a threads & queues solution that achieves a performance gain. And anything less than faster is a failure. Simply not worth the effort.

Not once. Not even your highly, (some might even say suspiciously), unrealistic dataset and duty cycle. Indeed, the more threads and queues you throw at the problem, the longer it damn well takes.

I'll let others decide for themselves about who followed a more realistic test scenario in the absence of any specifics about the OPs actual data and processing--which I asked for but never received.

Bottom line: A threads & queues solution will never solve the OPs problem. And I seriously doubt, but don't have the expertise to know, nor your code to try, that a forks & pipes solution will achieve much either. Assuming realistic datasets and duty cycles.

And the whole point of my posting some timing info was to force you to try and counter it. So the next time you get tempted to answer the OPs question with an untried, mind's eye solution:

Create the threads first and then have each thread load just the data it needs (and don't share it, of course). Then there won't be extra copies of that stuff created.

And only when pressed for more detail, did you 'fine tuning' that piece of off-the-cuff, based-on-nothing-but-what-you'd-read-somewhere, nonsense with more guesswork:

Having the parent read in the data and hand off each piece to the appropriate thread(s) (I'm guessing via Thread::Queue might be a good way) is the most general method that springs to my mind. I'd probably do something similar except using processes and simple pipes, as I've often done.

If you have neither the experience to know, nor the time to test your theories a little first, maybe you'll think twice, and so avoid wasting the OPs time pursuing impossible 'solutions'.

That's me done. Last word is yours.

---

Examine what is said, not who speaks -- Silence betokens consent -- Love the truth but pardon error.

"Science is about questioning the status quo. Questioning authority".

In the absence of evidence, opinion is indistinguishable from prejudice.

But replacing every character of every line is ... let's just say extremely unrealistic for now. So let's see what happens when we use what is only a very slightly more realistic dataset with a 50% duty cycle:

The work units are indeed quite small (1.0ms). But why then did you go and make the work units even smaller (0.5ms)?

But, and here is the real point. Never in any of these tests, yours or mine, your dataset or mine, have any of us found a threads & queues solution that achieves a performance gain. And anything less than faster is a failure. Simply not worth the effort.

Am I misreading your results? I see

• Not threaded = 23s
• T=1 = 25s
• T=2 = 13s
• T=3 = 10s
• T=4 = 9s

Huge gains except for a small penalty for T=1.