Explain intersection based vs. loop based rail networks

[Gus_Smedstad]

For that layout you can enlarge the whole intersection, making some internal blocks more than one train lenght

The problem isn't the length of the internal blocks, the problem is that the rails are close enough that you can't signal some internal blocks at all, regardless of the intersection size. The only real solution is to space the rails farther apart, either in general, or just before the intersection. The problem with the latter is that it's impossible to shift rails just a little - an S-shape shift shifts the rail a full 5 tiles over.

Earlier you said "at least a 4 tile gap is vital." It runs out you can place signals with a 2-tile gap, even though it looks like you can't. So a 2-tile gap isn't "way too close." But you can't place signals if there's no gap at all, which was true of one rail entering my intersection in my original screen shot. I've taken out one rail piece to make clearer that the end of a single curved rail is already intersecting the adjacent rail if there's no gap.

Also worth noting: in the first example, if I put a train stopped at that signal on the curved rail, and run another train into the intersection - they collide, causing damage. Not a problem with chain signals, but interesting anyway.

[hitzu]

The problem isn't the length of the internal blocks, the problem is that the rails are close enough that you can't signal some internal blocks at all, regardless of the intersection size. The only real solution is to space the rails farther apart, either in general, or just before the intersection. The problem with the latter is that it's impossible to shift rails just a little - an S-shape shift shifts the rail a full 5 tiles over.

I know, you're right. But either you can use tight layout with the possibility to block itself, or the wider one that is 100% safe.

Earlier you said "at least a 4 tile gap is vital." It runs out you can place signals with a 2-tile gap, even though it looks like you can't. So a 2-tile gap isn't "way too close." But you can't place signals if there's no gap at all, which was true of one rail entering my intersection in my original screen shot. I've taken out one rail piece to make clearer that the end of a single curved rail is already intersecting the adjacent rail if there's no gap.

If that works then it's worth to be used The only problem I see is the impossibility to make a crossover since they demand a 4 tile gap at minimum.

[ssilk]

This is a nice problem. I found an easy solution: double headed trains needs to be always an odd number of wagons. 1, 3, 5 wagons (more don't make sense). In that way, you can keep them symmetrical!

I don't get why an even number of wagons makes them less symmetrical.

Eh..... you're right, that was bullshit... I was so deep in my thought how I could make my 3 wagons supply train not dependent on direction he comes in, that I didn't saw thought about taking 4. Call it "focused on more important stuff".

Earlier you said "at least a 4 tile gap is vital." It runs out you can place signals with a 2-tile gap, even though it looks like you can't. So a 2-tile gap isn't "way too close."

As a rule of thumb I found, that 3 rails space between everything is enough to handle every situation. And it enables to put a rail between. If I think, this should be something like a "very big main line" I make a gap of 5, cause it enables me to put another 2 rails between, where the train can switch on (before crossings to the left or coming from the right), so that I have a buffer. It allows me also to put the cables between the tracks. An odd number here makes sense, cause you need one track gap to "switch a line" and if you have parallel lines in one direction you need also one track gap to place signals.

[sillyfly]

Trying to conclude the answer to the original question - we actually have two almost entirely independent aspects -
1. Single-headed vs. Double-headed trains (aka looped stations, but I *won't* be using this name, as it may cause confusion)
2. T-junctions vs. Roundabouts.

The first point has to do mostly with station design and expandability, while the second concerns only network expandability (and presumably efficiency).
So let's look at the two aspects in more detail:
1. Single-headed train-
Pros

• Simple to make and understand - you need one locomotive per train, and you always know which way it will face.
• Slightly shorter trains (for the same capacity). This is more apparent if you use shorter trains, obviously.

Cons

• Stations must be looped, that is - have an entrance and an exit. This makes stations larger and a bit harder to maneuver belts around, and also use a bit more rail segments (Unless you use siding stations, but they also force you to use roundabouts or else have external U-turns).

Double-headed train

Pros

• Stations can be smaller, and therefore need less planning ahead to expand. This makes double-headed systems better if you don't know how much you are going to expand, and wish not to reserve a lot of room in advance.

Cons

• You need double the locomotives, and therefore double the fuelling arrangements.
• Trains are longer (by exactly one locomotive each).

As for the claims that it is easier to get from anywhere to anywhere with double-headed systems - as far as I'm concerned these are still unbased claims. If anyone has a good reasoning as to why this is indeed the case I'd be happy to hear it.

No for the second part.

2. T-junctions

Pros

• Slightly smaller footprint.
• Uses slightly fewer rails for each junction.

Cons

• Each junction uses a few more signals (At least in designs I have seen).

Roundabouts

Pros

• Uses a few less signals for each junction.
• Trains can take U-turns.

Cons

• Junctions are slightly bigger.
• Junctions use a little more rails each.
• Trains can take U-turns.

The reason I put U-turns as both a pro and a con is that depending on the case they may cause troubles, but if you manage your train routes wisely (i.e. - not make trains take ridiculous routes) this is not a problem. Come to think of it, in T-junction based system the same problem would mean the train either finds a system-loop to use to take a U-turn, or else stops with "No Path" error, so I won't give it too much weight against roundabouts.

As discussed before - routing-efficiency-wise the two methods are identical.

Now, for the point of network expandability - in both cases you should have a blueprint of a 3-way junction, and plop it wherever you wish to make a new intersection. In both cases simply plopping the blueprint will work with no adverse effects on routing efficiency, deadlocks, etc. The only difference is that in the Roundabout method you will end up with rails going through the middle of the roundabout - this is not a problem as the signals will prevent deadlocks and maintain original routing efficiency. The only thing to consider is that you end up "wasting" those rail segments if you wish to leave them there. But then again, if you have a massive rail network stretching thousands of tiles, a few lost rail pieces probably won't matter that much.


Conclusion
In both cases it is mostly a matter of personal preference, unless you are extremely tight for space (in which case you are probably better off using belts instead of trains anyway).
T-junctions vs. Roundabouts - use whichever you like best, while considering that roundabouts will take slightly more space and rails, but T-junctions will take slightly more signals.
Single-headed vs. Double-headed - use whichever you like best, while remembering that double-headed systems are simpler to extend ad-hoc (that is - without prior planning), but need double the locomotives and longer stations.

You can obviously mix-and-match between the systems, having double-headed systems with roundabouts, or single-headed systems with T-junctions, or use a combination of double- and single-headed trains and T-junctions and roundabouts. Be careful though that if you have both double- and single-headed trains you must have exclusive stations for each type.

[Gus_Smedstad]

Trains are longer (by exactly one locomotive each).

And much slower, apparently. Though a really comprehensive analysis would have to look at the real increase in travel time, vs. just displayed speed, and consider how much time is spent waiting in stations vs. moving from station to station. Which of course depends on the rail network, what's being loaded, and how it's being loaded. The speed issue I imagine really crops up in post-victory mega-bases with long travel times between distant outposts and the main base.

[hitzu]

Trains are longer (by exactly one locomotive each).

And much slower, apparently. Though a really comprehensive analysis would have to look at the real increase in travel time, vs. just displayed speed, and consider how much time is spent waiting in stations vs. moving from station to station. Which of course depends on the rail network, what's being loaded, and how it's being loaded. The speed issue I imagine really crops up in post-victory mega-bases with long travel times between distant outposts and the main base.

The train capacity is so huge in terms of the amount of transported stuff that the little difference in the speed doesn't change almost anything. I use 1-2-1 and 1-4-1 setups and they cover all my needs even if the loading lime is set to the maximum, causing my trains just wait with no purpose half of their time. I tried to use 1-6-1 setups and desite the great speed loss they increased the capacity of my rail network.
I play with RSO mod, so the distances my trains have to travel are quite big.

[MeduSalem]

...

I have an additional point to make about the Pro's of Single-headed train stations and Cons of Double-headed train stations:


In reality railway companies used terminal stations in the past because of how during the history of industrialism many cities grew rapidly beyond their borders and the railway companies thought it to be much easier to just place terminal stations as close to the city as possible.

But nowadays this is considered a major conceptional disadvantage because ever since the amount of trains on the railway networks has increased vastly the terminal stations have become a huge bottleneck in the entire infrastructure because the trains have to enter and leave the stations through the same side.

That means that either on entrance the Trains have to cross all tracks, including the exit track, or vice versa, when exiting they have to cross the entrance track at some point. This results in a huge bottleneck at a certain traffic volume at which adding further parallel platforms at the station doesn't improve the capacity of the station anymore because either the leaving trains have to wait for the arriving trains to reach their platform or the other way around, that the arriving trains can't enter as long there's a train exiting a platform.

That means terminal stations are nowadays considered a huge no-go if they are meant for huge throughput with multiple parallel platforms and you'll see that many railway companies will invest billions to change them into through-stations because they offer vast improvements.

Not only can trains enter from both sides into the station enabling more overall routes, but it also increases the throughput for the station in one direction because of how a train may leave the station while another arrives at the same time on a parallel platform, without blocking each other in the process. This may increase the throughput by 2-4 times, depending on design.

This can either increase the overall capacity of the station or means that the same throughput as a terminal station can be achieved with a through station that has a much smaller footprint, requiring less parallel platforms.

The downside is that for trains that terminate in that station you will have to make a loop going around the entire station so that they don't have to leave the station through the same side they came from to avoid the crossing of entrance/exit tracks.


What this means for factorio is:

Single headed trains with through-stations are vastly MORE efficient than Double headed trains with huge terminal-stations.

Which is why I think that single-headed trains win over double-headed trains, because of how bigger main stations are possible without a decrease in efficiency.




Also for the roundabout VS T-junctions:


Roundabouts can have more than 3 directions. They work pretty fine for up to 8 directions without creating a huge track/signaling mess. The througput/efficiency for that is up for debate though, but never the less they offer more directional choice. You'd have to build several T-junctions to achieve the same, which often can't be afforded if there's not enough space. At that point a roundabout, even if it requires more initial space is still more space efficient.


[edit]Some small spelling errors and stuff.

[Gus_Smedstad]

Roundabouts can have more than 3 directions.

How often do you need that, though? 3-direction junctions, whether implemented as T-intersections or rotaries, are by far the most common type. I do find that occasionally I'll have a couple of 3-direction junctions near each other than might be converted to a 4-way junction, but the original concept was unrelated junctions.

After my recent experiences I'm inclined to favor rotaries over intersections, but I like to keep "typical use" in mind when discussing the advantages and drawbacks.

[sillyfly]

@MeduSalem - Thanks, I didn't even think of that.
Maybe there could be made a case for a hybrid system, in which you use double-headed trains with terminal stations at outposts (where traffic is much lower), but loop-based main stations. Of course, extra care should be made in such scenarios with signalling to avoid catastrophic failures.

As per the point about roundabouts able to easily have up to 8 exits - true, but I have purposefully disregarded this because, as Gus_Smedstad said, the most commonly used junction (in Factorio) is a three way.

[MeduSalem]

How often do you need that, though? 3-direction junctions, whether implemented as T-intersections or rotaries, are by far the most common type. I do find that occasionally I'll have a couple of 3-direction junctions near each other than might be converted to a 4-way junction, but the original concept was unrelated junctions.

After my recent experiences I'm inclined to favor rotaries over intersections, but I like to keep "typical use" in mind when discussing the advantages and drawbacks.

Well I have at least 2 spots in my railnetwork that require 4-direction roundabouts in my latest map and I've not found a viable way to get around them since I simply can't split them up in two T-intersections each due to lacking the space. Though they are not a real bottleneck so I don't care yet. xD

But yeah, the further away from my mainbase the easier it is to implement just a T-intersection or Y-intersection, depending on if I need to go in more than one direction. So you are particularly right on the fact that T-junctions are probably the most common.

It's just that if for some reason you need to add the 4th direction you can't convert a regular T-intersection into a 4-way intersection without tearing it down, which renders roundabouts to be the better default choice because you can easily add the missing direction later on.

@MeduSalem - Thanks, I didn't even think of that.
Maybe there could be made a case for a hybrid system, in which you use double-headed trains with terminal stations at outposts (where traffic is much lower), but loop-based main stations. Of course, extra care should be made in such scenarios with signalling to avoid catastrophic failures.

Well yes, that's perfectly possible if there's only 1 train that wants to visit a station at all times. Otherwise you need additional waiting blocks outside the station.


Currently I'm using a huge centered through-station as main station with 2x4 platforms (4 platforms for each direction) which can be entered from both sides, so it can handle 2 arriving and 2 departing trains at once without problems, while 4 are loaded/unloaded. Yeah it might actually be completely overkill xD

To be able to access/leave from both sides I've a huge bidirectional-ring enclosing my entire base in an octagon-shape (my defense follows that Octagon)... and the centered through-station connects with it on the East and West.

To get to the outposts I'm splitting off from the ring that encloses my mainbase and only use T-intersections the further away I am.

For the station at the outposts itself I'm using 2 platforms and a loop so I'm able to deal with one arriving and one departing train in parallel. This way I avoid that there might be a train waiting on the maintrack because it can't enter the station.

[sillyfly]

That sounds interesting!
Do you have images or videos of that base? Or better yet - a version of the save file somewhere online?

[MeduSalem]

That sounds interesting!
Do you have images or videos of that base? Or better yet - a version of the save file somewhere online?

Only a picture of an earlier stage from last weekend where I restarted to check out the 0.12 features for real... Didn't get to expand on it much ever since Wednesdays' Windows 10 upgrade though because I had several computer issues to deal with, also Prison Architect Alpha 35 took away some playtime too (yeah I'm cheating on Factorio, evil!):

Octagon base 1.jpg (64.03 KiB) Viewed 6825 times

The original base is somewhere off to the North-East. Also I haven't completed the ring entirely because I was still mining that huge patch of Iron ore that got in the way. So I had to temporarely use that little loop to make the rail work. Factorio... no matter how well you plan ahead the ore patches always get in the way at some point.

One thing I've done since is changed the upper station names into Main 2 or so to avoid that a train might want to go through the entire main station from East to West, then turn around at the roundabout and then head back and enter one of the other platforms. Happened to me already, but I fixed it. It's also the reason why there are only 6 stations on the picture instead of the 8 planned. xD

So I'm probably planning to use the 4 platforms called "Main" for the South/East directions and the other 4 platforms called "Main 2" for the North/West directions. That way the the trains won't interfere as much with each other on the roundabout.

[Gus_Smedstad]

Currently I'm using a huge centered through-station as main station with 2x4 platforms (4 platforms for each direction) which can be entered from both sides, so it can handle 2 arriving and 2 departing trains at once without problems, while 4 are loaded/unloaded.

I don't find myself building single central stations. Rather, I build a special-purpose station for processing each kind of cargo.

I'm probably not as far along, though. I just now found myself building my second copper train because my copper smelting station was outstripping the ore supply. Since each train is a 2-wagon train, I'm not really moving that much compared to what some are saying in this thread. I'm not even sure I would need 2 trains if I hadn't converted to double-headed trains for the copper supply line only, just to see how it worked. If I'd stuck with a single-headed train I probably wouldn't need a second one.

I'm thinking about either trying out one of the more extensive mods or moving on to another game. I feel like I've seen everything new in 0.12 now. I'm long past the rocket victory up to follower count 19 or 20. Which is I don't know how many actually followers, 50+ I'm guessing.

[Gus_Smedstad]

no matter how well you plan ahead the ore patches always get in the way at some point.

ALL the ore must be mined!

I also find it difficult to simply ignore ore patches when building.

It'd be interesting to find a need to build such a huge multi-platform station, but so far it'd just be pointless overkill.

[MeduSalem]

I don't find myself building single central stations. Rather, I build a special-purpose station for processing each kind of cargo.

[...]

It'd be interesting to find a need to build such a huge multi-platform station, but so far it'd just be pointless overkill.

Yeah it's basically pointless overkill and much more like a proof of concept to push the limits and nothing that's actually required... yet.

So it's because of the way I'm going to store & process ore altogether. I'm trying to build a smart furnace setup (including beacons) with about 30000-40000 Plates/Min peak production rate, with the option to double it. Yeah, ~8 full chests of plates per minute. That thing wants to be fed appropriately. Quite ridiculous but I like it that way.

I basically unload all ore/coal/stone from the trains into active provider chests, then store everything into the storage chests (don't have that many yet, only the one blue square near the train station) or directly to the requester chests of the furnaces and process them into plates for further storing. The entire thing is trapped inside a 2x2 roboport subnetwork, so that the robots can't leave it.

So much about self-set goals. xD

ALL the ore must be mined!

I also find it difficult to simply ignore ore patches when building.

Well I can't ignore resource patches either. It's like a thorn that must be removed. So much greed. xD

[Linosaurus]

Yeah it's basically pointless overkill and much more like a proof of concept to push the limits and nothing that's actually required... yet.

You know what would make big train stations more useful? If train wagons had a *lot* less capacity.

... can this be modded?

I'm actually tempted by the idea.

[Gus_Smedstad]

I was refactoring part of my rail network today, and I ran into something I was vaguely aware of before. Namely, that for a minimum-radius rotary, the rails entering the rotary must be at least 90 degrees apart. Attaching to points 45 degrees apart makes rails cross. Thus a minimum-radius rotary can only service 4 rails, not 8.

Increasing the radius should allow for more connections. It also follows that you can connect more than 8 if the radius is big enough - there's no upper limit as long as the junction keeps getting bigger, though obviously it gets more and more inefficient.

I have a point where 5 rails come together in somewhat close proximity. At first I had that as a Y and two T junctions, but I needed to convert the Y to a T to allow trains leaving a station to turn left, and I couldn't place two T junctions in such close proximity. I refactored it as two rotaries and a T, but I'm thinking about revisiting it to make it a single rotary.

[OBAMA MCLAMA]

Can you show us a 4 lane and 8 lane loop then gus?

I would also prefer you to do heavy testing on it also but I mostly just want to see the loop/roundabout/rotary.

To increase throughput, some of us have started on 8 lane rail systems.

[Gus_Smedstad]

Can you show us a 4 lane and 8 lane loop then gus?

I'll take a stab at that.

I looked at my 5-way intersection, and decided that 2 of the rails were far enough from the major intersection to justify leaving them as an existing T-junction.

The Problem

I did, however, consolidate the two rotaries into a single rotary. I realized that I could have gone with a conventional minimum-radius rotary if I changed the angle where one rail entered, but I wanted to see what it looked like if I forced a 45-degree separation between entries. This is as small as I could make it.

45 degree rotary

[Gus_Smedstad]

OK, I made a rotary with 8 connections. It's exactly twice the diameter of a 4-connection rotary. Also, I was forced to use minimum spacing between the rails for entering / exiting connections - wider spacing requires a larger rotary.

I've included a 4-way rotary for size comparison.

8 way rotary