Timeline for How does variance in task completion time affect makespan?

8 events

when toggle format	what		by	license	comment
May 7, 2013 at 16:21	vote	accept	Patrick87
May 7, 2013 at 9:21	comment	added	András Salamon		I don't know where the "hunch" came from; it is not consistent with the rest of the heuristic reasoning.
Apr 19, 2012 at 20:15	vote	accept	Patrick87
May 7, 2013 at 16:21
Apr 13, 2012 at 22:13	history	edited	Gilles 'SO- stop being evil'	CC BY-SA 3.0	formatting
Apr 13, 2012 at 14:36	comment	added	Patrick87		the first is justified by the fact that $1 - (1 - P(X = 5)^k)^n$ tends to $1$ for fixed $k$ and increasing $n$; the latter by the fact that $Var[X + X] = Var[X] + Var[X] = 2\sigma^2 \leq 4\sigma^2 = 4Var[X] = Var[2X]$... so the variance doesn't increase linearly as a function of $k$. Is that compatible with your thinking (that's how I'm interpreting what you have so far)?
Apr 13, 2012 at 14:34	comment	added	Patrick87		+1 Excellent intuition, and this helps to clarify my thinking as well. So increasing processor counts tends to increase makespan under a weak scaling assumption; and increasing task counts tends to decrease makespan under a strong scaling assumption (of course it takes longer; I mean the work/makespan ratio improves). These are interesting observations, and they seem true;
Apr 13, 2012 at 14:22	history	edited	svinja	CC BY-SA 3.0	added 228 characters in body
Apr 13, 2012 at 14:14	history	answered	svinja	CC BY-SA 3.0