Estimated number of API servers under high concurrency

My main job is to write the API, so how to test the API response time, as well as the expected high concurrency server pressure?

1 The total number of API daily visits is first counted, for example a.

API generally has access logs, statistical access logs. Linux uses a similar grep WC pipeline command statistic.

2 computes the number of concurrent numbers.

How does this count? Estimate it. Here we have a principle: 80% traffic is concentrated in 20% time.

m = a*0.8/(24*0.2*3600) Gets the approximate number of concurrent m per second.

3 assume that the agreed API response can endure a maximum of B MS and test the number of concurrent times when the response time is b MS

Linux under the AB performance test, as follows (this oneself slowly try out)

Ab-c 100-n 10000 http://mixservice/shop/template?shop_id=15141

-C is the number of concurrent,-n try to set a larger.

Get the critical concurrency C when the response time is B Ms.

4 assuming 5 times times the pressure, how many servers should be needed.

5M/C (assuming that C is concurrency that can be handled by a single server)

Estimated number of API servers under high concurrency