Making Programs Faster

When It's Time to Give Up

     %Time ExclSec CumulS #Calls sec/call Csec/c  Name
      63.7   0.830  0.829    109   0.0076 0.0076  main::letter_histogram
      13.8   0.180  0.180      1   0.1800 0.1800  Mail::Util::read_mbox
      4.61   0.060  0.130      3   0.0200 0.0432  main::BEGIN
      4.61   0.060  0.887    109   0.0005 0.0081  main::handle_message

We could conceivably save up to 180 ms per run by sufficiently clever hacking of read_mbox
- How much is that pony really worth?
Say my computer cost $3000 and has a lifetime of about 5 years
- That's about .0019 cents per CPU-second
- The benefit of a 20% speedup in read_mbox is about .000000676 dollars per run
- That's the pony. What is the price?
My time bills at a fairly high rate, but let's say it's $50 per hour
- I might spend 20 minutes getting the speedup
To break even, I would have to run the program about 25 million times
Of course, this is much more likely if the program has 25 million users