I found that all say that throughput is inversely proportional to response time. But I found it's confusing that if I see their relationship by their definitions, isn't each of them a reciprocal of each other?
From my point of view, if I define them by my intuition, it then goes:
Throughput: Number of task completions per time unit = $\frac{Completion}{T}$
Response Time: time cost from task's arrival to its completion = $\frac{T}{completion}$
where $T$ is the total observation time.
What did I miss? Why can't I say that $Throughput = \frac{1}{Response Time}$?
I think I've found my bug,
The mean response time should be $\frac{\text{Total Response time of all jobs}}{completion}$, and every job has a different $Response\ Time$ due to the queueing delay.
This is the main observation that it can't be the reciprocal of throughput, and especially in the parallelism case.