scalability - Calculate load for performance testing from requirements

Question

Suppose we have an API endpoint that will receive 100 Million requests a year and suppose our application needs to have at least 95% of those requests handled with less than 30ms. Is there a model to calculate what will be the load for the performance test that can make these tests check if the current system can meet those requirements?

With load, I mean, for instance: in the test, we need to do x parallel requests and have q% of them with less than t ms.

Answer 1

The "model" depends on your application usage patterns, i.e.

• whether it is used 24x7 or rather during business hours
• are there any spikes
• is the load distributed evenly
• what is the logic behind different API endpoints if you have > 1 because /health or /status will be very fast and some wildcard search or bulk edit operation will be much longer

In general 100 million requests / year is not that big amount,

• 100 000 000 / 365 ~ 274 000 requests per day
• 274 000 / 24 ~ 11500 requests per hour
• 11500 / 3600 ~ 3 requests per second

So you can use any load testing tool which supports:

• HTTP protocol
• including the ability to add HTTP Headers
• including the ability to report percentiles
• including the ability to apply pass/fail criteria so the test execution could be marked as failed if response time exceeds acceptable