Cost of dynamical quark simulations with improved staggered quarks

Several caveats should be pointed out to the reader. First, I think that the past 20 years of my life are evidence of my inability to estimate the time and effort necessary to calculate the spectrum of QCD. Used to doing analytic calculations that took a few months at most, I did not imagine as I started my first Monte Carlo calculation that 20 years later I would still be doing similar calculations. I was a postdoc at Fermilab then, and as I write this I am back at Fermilab on sabbatical. With great pleasure I see how far we have come, and look forward to an exciting future of improved calculations. Caveat two is that this write-up is not a transcript of what I said in Berlin. Caveat three is that this does not represent a MILC consensus statement. I did my best to extract from past experience what is required for future calculations, but the whole collaboration has not had an opportunity to check or react.