Comment by cortesoft | Hacker News Reader

Comment by cortesoft | Hacker News Reader

Parent Full thread

Some of the areas the article leaves out are around throughout and bottlenecks... you are often limited by particular resources, and the ability to get them to the end of your factory line (I.e. the belts are saturated at the start of the line, but empty by the time they get to the new section of the factory)

Another interesting thing you learn is about a form of 'buffer bloat' (although it behaves differently than traditional network buffer bloat)

Essentially, having too much of a buffer for resources can cause inefficiency.

Say, for example, that you notice you are producing more green circuits than you are using. A naive response is to try to store them, so that your green circuit factories are running at full speed, and then when your need for them increases you will have them available.

However, this has two downsides. One, it will pull resources away from things you actually need in order to keep making more green circuits, even though you have more than you can use right now anyway. The copper that could be going to manufacturing the item that is your current bottleneck is instead going towards something you have too much of anyway.

The second issue hits you on the other side... when you DO start using more green circuits than you are producing, you won't notice until your issue is very severe, because you will have been using your reserves for quite a while. By then, you will have to increase your production a lot to make up for it... which might put a different resource out of whack.

The trick is to let the factories overproducing to backup production and stop... let that back pressure behave as a natural production regulator, which will make it much easier to have a factory that always works as efficiently as possible given your current raw material production.

Yep, the only places that I ever really need a buffer are train stops, power facilities, and times when you can't use a byproduct just yet (more-complicated refineries before you get your circuits tuned just right).

One thing I just realized this week is that a power generation buffer needs to alert me on the very first swing of an inserter from a buffer chest. If the coal/solid fuel line isn't completely compressed, that means the doomsday clock has started ticking.

I recommend avoiding steam buffers when you go nuclear. I had a setup with multiple storage tanks to hold steam, and that had buffer bloat because fluid/steam is hard to evenly distribute across many paths.

To be efficient with nuclear, though, you need steam buffers... nuclear power can’t stop burning fuel once it starts, so unless unless you are spending all the power as you generate it, you need to store it. Burn a unit of fuel, store all the steam, and then don’t put more fuel in the reactor until the steam level goes below a threshold, then repeat.

This is especially true because you want to have a group of four reactors together for maximum efficiency, which generates a crazy amount of power. You spend more time not generating more heat and just living off the steam storage than you do burning fuel.

Ok, but it isn't worth it. Yes, you get more "power value" out of nuclear fuel if you ration it. But even if you overbuild your power by 2x your current target load and let the extra go to 'waste', your ore reserves will still likely give you many hundreds of hours of power.

I've tried and made many buffered, circuit-controlled, meticulously fine-tuned power plants that are optimized for fuel conservation and they always have some hitch or edge case where they act erratically, or the steam is not balancing, or the reactors cooled down too much, or... (Or you're lategame and you outscaled it too far and need to eke the UPS back.) And is the moment when you're under siege and losing losing power really the best time to be debugging that 30-combinator fuel-optimizing circuit network you picked up off reddit and randomly modified until it seemed to work? Right.

A challenge: calculate how many hours the total amount of ore you are willing to mine will last you. If you maximize efficiency through productivity modules -- even gradually -- I bet it will be hundreds if not thousands of hours.

So you make a trade: a small fixed increased cost in fully automating and maximizing nuclear fuel production in exchange for the benefit of vastly reducing the complexity of construction and operation of your primary power production facility. No tanks, no circuits, no buffering -- instead, full automation and productivity for fuel production and 100% power all the time. Give yourself a(n automated) backup chest or two of fuel nearby for days (calculate it!) of time to deal with fuel production problems. Would you even have that much time if you burned all the coal in the same area?

I would take a trade like that every day.

> A challenge: calculate how many hours the total amount of ore you are willing to mine will last you. If you maximize efficiency through productivity modules -- even gradually -- I bet it will be hundreds if not thousands of hours.

What do you mean "hours"? It'll last multiple real-time years.

>meticulously fine-tuned

On what planet is a bang-bang controller considered a fine-tuned control system in an automation game? The only simpler system is proportional.

The point of doing it isn't necessarily for some maximum efficiency, but because it is the only problem in the entire game that could even ask the player to make a control system, which is a very fun thing to do.

If you build a controller for a single nuclear reactor, a bang-bang controller is indeed pretty simple once you know how to build an SR latch.

Scaling that to multiple reactors becomes complex because you can no longer control reactors independently and still get adjacency bonuses from them operating together. Further, there are complexities from emergent behavior that you have to control for. Steam turbines, like steam engines, auto scale steam consumption to power demand. The problem is they don’t do it uniformly. If you have 2 turbines attached to 2 tanks and you only need 1, the game might consume all of 1 tank first, leaving you with the 50% burst capacity. It becomes a big problem with day/night cycles and solar/accumulators in the grid.

I never tried to balance this, but I suppose it’s possible by building a circuit to link together all tanks and shut off their flow when one‘s storage deviates too much from the average. This sounds really complex to me and not worth it.

I’ve explained to the best of my understanding of the game engine’s behavior, but I know that fluid mechanics is also inaccurate and complex to design for. There’s a known issue with flow behaving asymmetrically. One reason is because the game updates state in a certain order, so water and steam flow with a bias for certain directions (n, s, e, w).

Also, thanks for teaching me a term from control theory (1)

(1) https://en.m.wikipedia.org/wiki/Bang–bang_control

Ah individual control of a large array of reactors is trickier. I think the bang-bang controller with keeping all reactors synchronized has good legs, but it stops scaling well at some point. If the buffers become the limitation and you don’t want to run all of the reactors synchronized then a PID controller would be a good pick. The output value would choose how far up the row of reactors you enable. If you could force max load on will, then loop tuning is a possibility.

To maintain the simultaneous reactor bonus, you would want to keep the fueling cycle synchronized and just adjust which reactors are enabled each cycle. This can be done with a clock that has a period of one fuel cycle.

One interesting thing is that due to the nature of reactors, you will have higher efficiency in running more reactors if your buffer does not run out. It might be possible to eek out a little more efficiency if you fall back to a bang-bang controller in low load scenarios (making sure the threshold and number of reactors used won’t overfill the buffer, which is practically unlikely in low load scenarios). I would have to experiment with this a bit more to see if it’s reasonable to capture more efficiency.

All of this is practically orthogonal to the standard players perception of efficiency, but I find it fun.

https://en.wikipedia.org/wiki/PID_controller

https://github.com/quchen/articles/blob/factorio/factorio/ci...

Fascinating write up! Thank you for sharing!

I mean, it's a game, though. A lot of the things I do when I play games "isn't worth it." But I have a lot of fun, anyway. :)

Reactors don’t top out in efficiency at 4. I went from a 2x2 to a 2x7. My 2x4 power plant was buffered and would create flow problems when half of my buffers drained before the other half. When I scaled to 2x7, I dropped all but one tank to let me monitor when my reactors were above 500C.

My biggest learning with nuclear is that if I’m producing too much power for my factory, I’d rather grow my factory to meet my power supply than to make a more sophisticated power plant. Your mileage may vary.

Reply on news.ycombinator.com