Around this same time (1998) backoff thresholds on dampening and other issues with BGP convergence were benefiting from another recent and conceptually related

Around this same time (1998) backoff thresholds on dampening and other issues with BGP convergence were benefiting from another recent and conceptually related paper on how random_drop being configured on tcp ports prevented SYN flooding. Lately, we have ChaosMonkey and other random failure generators for uptime as well. The thesis has the underlying principle so well articulated though. Adding randomness to improve robustness is a really beautiful idea. What caught me is that you found the effect was a performance increase in the overall system as a result of the additonal entropy. Nassim Talib took it up later and made it famous, but it's a deep principle. I was wondering how generalized we could say if, that you have a process of any kind, which is subject to another random process that stresses it, and it evolves and adapts in response to the random stressors, and performs better as a result, could we find a general accelerating principle for a given system? As in, at a level of abstraction, is this the same principle at work as in a random neural net? (Edit: this was just posted to HN frontpage, which runs down this rabbit hole: https://www.quantamagazine.org/anil-seth-finds-consciousness-in-lifes-push-against-entropy-20210930/ https://www.quantamagazine.org/anil-seth-finds-consciousness... )