alt Hacker NewsIt strikes me that a problem with this analysis is with the modeled example system itself. The assumption that stock should follow immediately preceding demand is valid but typically wrong.
Pursuing the intellectual mechanics of a system is obscuring the error of the wrong system!
Instead stock levels should take into account lead time, historical seasonal demand, and a subjective value. The observation and order frequency should be in months not days. While computing allows near “real time” everything, that doesn’t mean it should necessarily be applied.
Just-in-time thinking needs to take the reality of the system (in this case the supply chain) and other factors like hedging opportunities?
Replies
It applies to software engineering too in things like autoscaling. I’ve seen many DDOS outages caused not by the attack itself but the wave of autoscaling that takes things down (often well after the actual attack ended).
It’s a simplified example of a broader principle. This structure shows how oscillation can emerge from a very simple system and how it can be remedied by adjustments of the delays. Oscillating systems are very much a thing and this explains why.
By the way, what you said about stock levels is correct, of course, but doesn’t necessarily avoid oscillations. See, all of these metrics are feedback-based as well. With a delay. How much history do we take into account? Are account lead times influenced by delivery times. This is a classic problem in system dynamics: where do we draw the system boundary? Since everything is connected to everything, the right thing to do is to model the entire universe. Since this is not practical, we draw a line _somewhere_. _Where_ exactly does not matter much as long as the resulting model is useful.