The servers are overloaded, and everything grinds down to a sluggish horror. What now?
If you’re able to pinpoint performance problems down to a single or a few request types, you can try to manipulate the percentual share of these requests in the overall requests. The goal is to reduce the amount of requests to the trouble makers.
This can be achieved by adding a sleep command into the requests responsible for the slow down.
That way you can reduce the throughput of these requests, worsening their response time even more. But all the other requests can gain ressources, and their throughput can raise.
Downside
- The result heavily depends on the reason why the trouble making requests are called.
- Disabling the feature or serving an error 50x instead maybe more honest, and much more effective.
- Finally, this is NOT a solution at all. It’s a temporary help until you have a solution.