Adding processors does not make a fixed workload proportionally faster when some of its work must remain serial. Amdahl's Law quantifies that limit.
Design an AmdahlSpeedupCalculator class:
AmdahlSpeedupCalculator() creates a stateless calculator.double speedup(double parallelFraction, int processors) returns the speedup over one processor, rounded to 5 decimal places.
Treat the original runtime as 1:
- The serial fraction is
1 - parallelFraction and does not speed up. - The parallel fraction takes
parallelFraction / processors time. - The new runtime is
(1 - parallelFraction) + parallelFraction / processors. - Speedup is
1 / newRuntime.
Round only the final speedup. Each call is independent and must not use values from an earlier call.
Example 1:
Input:
Output:
Explanation: The new runtime is 0.1 + 0.9 / 4 = 0.325. The speedup is 1 / 0.325 = 3.076923..., which rounds to 3.07692.
Example 2:
Input:
Output:
Explanation: The new runtime is 0.05 + 0.95 / 8 = 0.16875, so the workload is about 5.93 times faster rather than 8 times faster.
Constraints
0 <= parallelFraction <= 11 <= processors <= 10^6- Round the final answer to 5 decimal places.
- Answers are accepted within
10^-5 of the expected result. - At most
100 calls are made to speedup.
Calculate Amdahl's Law Speedupeasy