Pipelining and Parallelism
Goal: trading computation for (gain in) time.
Degree of parallelism: response time x throughput
Linear speedup: split up task into N independent subtasks, each requiring same amount of time.
- Response time speedup of N. Throughput constant. Degree = N
Pipelining: Can't independently split subtasks - the subtasks may be serially dependent.
- We can get speedup in throughput, but NOT in response time by using pipelining