Next: AMDAHL'S LAW
Up: Evaluating Speed-up Mechanisms: limits of parallelism
Previous: Evaluating Speed-up Mechanisms: limits of parallelism
Suppose you invent some clever trick (cache, pipelining, special
instruction, parallelism)
- 1.
- How good is it?
- When the trick is used, what speedup over the unenhanced case is achieved?
- 2.
- How much good is that?
- For what fraction of the overall execution time can the trick be employed?
So what's the overall benefit?