Describe the components of the Amdahl's law equation. (Btw, what type of scaling is this speedup calculation assuming?)
funkysenior15
@kayvon: I'm not quite sure what you mean by "components". But is this speedup calculation assuming problem-constrained scaling?
kayvonf
I just meant the parts of the equation. And yes, we are talking about problem constrained scaling here.
amaliujia
This is problem constrained scaling I think, as Amdahl's law tries to estimate upper bound of speed up we can get, that is, upper bound of how fast we can solve problems.
Component:
f is fraction of parallel part of program. If f is 1, then all thing in program can parallel, we get n speed up(the best :)). If f is 0, then nothing can be parallel, so no speed up. To summary, assume n is fixed, the larger f, the bigger speedup we get.
n is number of cores. So assume f is fixed, the more n, the faster program run.
So Amdahl's law tells two factors constrain us achieve perfect speedup
skywalker
(1-f) = fraction of the program that is strictly serial
f = fraction of the program that can be run in parallel
Describe the components of the Amdahl's law equation. (Btw, what type of scaling is this speedup calculation assuming?)
@kayvon: I'm not quite sure what you mean by "components". But is this speedup calculation assuming problem-constrained scaling?
I just meant the parts of the equation. And yes, we are talking about problem constrained scaling here.
This is problem constrained scaling I think, as Amdahl's law tries to estimate upper bound of speed up we can get, that is, upper bound of how fast we can solve problems.
Component:
f is fraction of parallel part of program. If f is 1, then all thing in program can parallel, we get n speed up(the best :)). If f is 0, then nothing can be parallel, so no speed up. To summary, assume n is fixed, the larger f, the bigger speedup we get.
n is number of cores. So assume f is fixed, the more n, the faster program run.
So Amdahl's law tells two factors constrain us achieve perfect speedup
(1-f) = fraction of the program that is strictly serial
f = fraction of the program that can be run in parallel