Adaptive http request algorithm

I am currently working on a splash based backend project. if you don't know what a spray is, it doesn't matter, just rate it as an HTTP request handling system. unfortunately srpay does not have existing support for throttling requests. so I would like to write my own.

I don't want to use a token or similar algorithm as the server capacity is preconfigured. you can make a very conservative estimate that lags far behind the actual server capacity. so what I would like to do is let the server actually examine its own capacity using request per second feedback, request per second. processing requests per second and average response time. Usually four statistics, but not limited to.

This is adaptive throttling, so the system dynamically knows the actual bandwidth of the request. Anyone can suggest some existing algorithms or related documents.