Is Amdahl’s law a friend or foe of parallel processing?
What will be an ideal response?
Although Amdahl’s law allows you to estimate how much improvement parallelizing a problem will give you, some commentators have stated that it has throttled the development of parallel processing by saying that the level of improvement you can get is very much related to the fraction of the program that cannot be executed in parallel. Because Amdahl’s law says that the ultimate speedup is limited, some researchers have felt that there was no point in pursuing high levels of parallelism. Why spend a lot of effort chasing ever increasing levels of parallelism if it leads to a dead end? In that sense, Amdahl’s law has acted as a brake on progress. The more cheerful Gustafson’s law says that as problems grow (i.e., are scaled up) the number of operations that can be parallelized increases and the benefits of parallelization continue to grow. In other words, Gustafson is saying that the ratio of serial to parallel processing time is decreasing as program complexity increases.
You might also like to view...
Answer the following statements true (T) or false (F)
1) A key to improving the software development process is encouraging software reuse. 2) To enhance performance and reduce errors, it's a good idea to make derived classes larger than they need to be. 3) A base class is often designed by factoring out commonalities among a set of classes. 4) When creating derived classes, you must use discretion in choosing the proper base class. Ideally, the base class will not contain excessive capabilities or information. 5) Method Equals will accurately compare any two objects.
You use a(n) ________ control to display text or decorative elements on a form
A) freeform B) bound C) unbound D) graphical
The appropriate format is to display the header of a two-page letter on both the first and second pages.
Answer the following statement true (T) or false (F)
The function call ____ passes a copy of the complete emp structure to calcNet().
A. calcNet(struct emp); B. calcNet(*emp); C. calcNet(&emp); D. calcNet(emp);