> I agree with his rant about "CS departments' algorithmic analysis" failing to cover these real-world hardware issue. That was certainly true of my ~1997 under

> I agree with his rant about "CS departments' algorithmic analysis" failing to cover these real-world hardware issue. That was certainly true of my ~1997 undergrad CS course at the University of St Andrews, that was otherwise exemplary. I counter with similarly anecdotal evidence from 2012. :) While we had a normal "Algorithms and Datastructures" course that taught the classical theory, there was also a separate course about "Algorithm Engineering" which was (among other things) about performance in the presence of the memory hierarchy. The upshot is that there are different cost models that you have to use in the analysis of your algorithms (e.g., IO complexity), but the techniques that you use for the analysis are the same. --- To be honest, I think the system is working as intended here. Undergraduate courses are not really designed to push people to the forefront of current research. They're intended to give you the basic skills you need to catch up to modern developments yourself. I think what's happening here is that people remember "optimality proofs" from their algorithms courses and take that to mean that there is nothing more to be done (that's the sentiment I got from this blog post anyway). The problem with this is that in this context "optimality" referred to a technical definition - optimal with respect to some specific simplified model of computation. The result is probably not optimal with respect to a more complicated model!