3 ms·
Control systems, yes, but we have to learn to crawl first
- thenewwazoo 7y agoWhat a weird post. I actually just gave a talk at an internal conference about an early-days PID control loop for doing capacity management (and am generating gain parameters as I type). Is her argument that I shouldn't have built it because... the people who will benefit from it don't know how to use it? Or something? Not every problem can be solved by implementing a controller but lots of them can, and "it's risky!" seems like a poor reason to avoid working toward a control plane.
- singingboyo 7y agoYou have to read the previous post (linked by "Yesterday's Post" at the start of the article) and it makes more sense. In a system with as many issues as the one described there, a controller is probably not the best use of time - just increasing minimums at peak times would work fine, and there are other problems that are probably more important.
- thecopy 7y ago> the people who will benefit from it don't know how to use it? Or something? Something like that. Another point is that the complexity adn increased risk (yes) added by complex automated (auto-scaling-)systems might not be worth the price you save instead of running at 5x mean capacity all the time.
- gumby 7y agoGreat! None of this applies getting your system set up and running (the classic MVP). But eventually, if all goes well, you want to scale. Then the choice of "take the harder path" really pays off. The trick is to have at least a person or two on the team when you're implementing the MVP who are thinking about the long term architectural issues. You don't want to do a complete reimplementation.