Identifying positive feedback loops

Positive feedback loops can cause chaotic behavior and catastrophic reductions in performance, where adding load increases failures, which adds even more load:

Figure 13.4 – A positive feedback loop of loading failures producing even more load

Inefficient processing can be due to failures requiring more system resources, for instance, due to writing large error logs or remote calls to indicate a problem. Additional retries can be caused by faults in the client logic, such that some failures lead to even more requests and even more failures.

Recall the example of self-inflicted DDoS from Chapter 7, Testing of Error Cases. There, positive feedback resulted in more load and even more failures. We saw an example when we read conference details from Outlook for all our users: the system was overwhelmed, and writing the logs for all the failures only slowed it down further.

Often these failures are emergent behavior...