4-way intersections: Throughput and deadlocks [image heavy]

Smart setups of railway stations, intelligent routing, solutions to complex train-routing problems.
Please provide - only if it makes sense of course - a blueprint of your creation.
AlexAegis
Inserter
Posts: 35
Joined: Sat Jun 04, 2016 10:25 pm
Contact:

Re: 4-way intersections: Throughput and deadlocks [image heavy]

hansjoachim wrote:
Thu Jan 24, 2019 9:05 pm
AlexAegis wrote:
Thu Jan 24, 2019 5:52 pm
Sadly I don't know how can I improve this further
So I improved it with correct signals.
I also made a better version with extra right turns.

Original: 30 trains/min set 1

Correct signaled: 41 trains/min set 1

With extra right turns: 48 trains/min set 1
btw: those test are with rocket fuel. With Nuclear fuel the numbers would be higher
Thanks! Yes, I'm using nuclear fuel that's why my numbers are higher. I wanted to let the trains as far in into the intersection as I could without completely deadlocking it, that's why I don't have chain signals everywhere. Guess it was a bad idea. Also, on every left turn before merging back to the "output" track why isn't that signal a chain signal?

hansjoachim
Fast Inserter
Posts: 122
Joined: Wed Apr 26, 2017 7:03 pm
Contact:

Re: 4-way intersections: Throughput and deadlocks [image heavy]

AlexAegis wrote:
Fri Jan 25, 2019 12:22 am

Thanks! Yes, I'm using nuclear fuel that's why my numbers are higher. I wanted to let the trains as far in into the intersection as I could without completely deadlocking it, that's why I don't have chain signals everywhere. Guess it was a bad idea. Also, on every left turn before merging back to the "output" track why isn't that signal a chain signal?

Nuclear explains it partly, but not from 30 to 50.
Left turns are often the limiter so if you have more straight turns or right turns your number would be higher. But also if all trains go left your numbers would be higher as every cycle two trains can pass the intersection.
Your test isn't wrong, it's just different.
The new signals might not give an increase in your test

The last signal isn't necessary, you don't need chain signals for merging.
Updated version of aaarghas intersection testbench. viewtopic.php?f=194&t=46855&p=383830#p383830
My Highscore intersections: https://pastebin.com/dGsJT71Z
My 3 to 1 train merger https://pastebin.com/AcLHXcwd

Dolu1990
Burner Inserter
Posts: 5
Joined: Mon Feb 11, 2019 12:00 am
Contact:

Re: 4-way intersections: Throughput and deadlocks [image heavy]

I got a 2 way, 415 trains on 5 minutes => 83 trains/minute (3 length trains)

This solution has some good sides :
- Trains can turn back by crossing the middle section three times.
- It can be scaled without too much cooking
- The flow is good, the main slow down come from the middle horizontal vertical 6x6 cross

Deadlock/congestion of the loops is avoided via wire signals.
intersection
screenshot.png (916.94 KiB) Viewed 1331 times

hansjoachim
Fast Inserter
Posts: 122
Joined: Wed Apr 26, 2017 7:03 pm
Contact:

Re: 4-way intersections: Throughput and deadlocks [image heavy]

Dolu1990 wrote:
Mon Feb 11, 2019 12:34 am
I got a 2 way, 415 trains on 5 minutes => 83 trains/minute (3 length trains)

This solution has some good sides :
- Trains can turn back by crossing the middle section three times.
- It can be scaled without too much cooking
- The flow is good, the main slow down come from the middle horizontal vertical 6x6 cross

Deadlock/congestion of the loops is avoided via wire signals.
Separating left, right and straight
Multiple lanes to increase throughput in the center.
Staying at 2 lanes
The signals are correct just lacking a couple.
The throughput is significant higher than the best unbuffered intersections.
It also looks good

I haven't checked out the circuitry but do you need it? Well probably for u turn

This intersection is really designed for 2 levels, as everything is in one level in factorio the left turns doesn't make sense. They cross many too many paths. You should let left turn drive to the left first and not to the right first.
The U turn here is horrible as it crosses way too many train paths multiple times. You should have a good U turn before this intersection to prevent trains from taking U turns in this intersection.

There are some lacking signals in the cross where left turns trains "reenter". It should be seperated with signales to increase throughput.

To compare this to the other intersections and to get it added to the list someday, you need to make a 6 car design.

Tldr
Add U turns before the intersections
Add some signals to fix the 'cross'
Left turn is this intersection largest problem and should be addressed
Make a 6 car version aswell
Updated version of aaarghas intersection testbench. viewtopic.php?f=194&t=46855&p=383830#p383830
My Highscore intersections: https://pastebin.com/dGsJT71Z
My 3 to 1 train merger https://pastebin.com/AcLHXcwd

Dolu1990
Burner Inserter
Posts: 5
Joined: Mon Feb 11, 2019 12:00 am
Contact:

Re: 4-way intersections: Throughput and deadlocks [image heavy]

This intersection is really designed for 2 levels, as everything is in one level in factorio the left turns doesn't make sense. They cross many too many paths. You should let left turn drive to the left first and not to the right first.
When you say left turn, is it from a left hand drive or a right hand drive perspective ?
I mean on the picture, which is left hand drive, the left turn is the first thing the train can do, and it cross nearly nothing.

About wire signals, they are required to avoid having all the 24 inner slots busy, which would create a dead lock.

Dolu1990
Burner Inserter
Posts: 5
Joined: Mon Feb 11, 2019 12:00 am
Contact:

Re: 4-way intersections: Throughput and deadlocks [image heavy]

Were you talking about the left turn after the middle cross ?

hansjoachim
Fast Inserter
Posts: 122
Joined: Wed Apr 26, 2017 7:03 pm
Contact:

Re: 4-way intersections: Throughput and deadlocks [image heavy]

Oh I am sorry, its right turns for LHD
Updated version of aaarghas intersection testbench. viewtopic.php?f=194&t=46855&p=383830#p383830
My Highscore intersections: https://pastebin.com/dGsJT71Z
My 3 to 1 train merger https://pastebin.com/AcLHXcwd

Dolu1990
Burner Inserter
Posts: 5
Joined: Mon Feb 11, 2019 12:00 am
Contact:

Re: 4-way intersections: Throughput and deadlocks [image heavy]

Ahhh, ok ^^
How would you " You should let left turn drive to the left first and not to the right first" ? (LHD perspective)

Dolu1990
Burner Inserter
Posts: 5
Joined: Mon Feb 11, 2019 12:00 am
Contact:

Re: 4-way intersections: Throughput and deadlocks [image heavy]

Damned i mixed myself with LHD / RHD.
My question was, what exactly do you mean by "You should let left turn drive to the left first and not to the right first" (RHD perspective)
You mean before the middle cross ?

Themme
Burner Inserter
Posts: 7
Joined: Tue Jan 08, 2019 11:29 am
Contact:

Re: 4-way intersections: Throughput and deadlocks [image heavy]

Hello folks,

Ive been messing around with some intersections, and I am kinda curious what you think of this design, how to measure throughput?
Ive created it myself a moment ago in my own factorio save.

https://www.dropbox.com/s/vhn8s5as89ipu ... 1.jpg?dl=0

Thanks, Themme

hansjoachim
Fast Inserter
Posts: 122
Joined: Wed Apr 26, 2017 7:03 pm
Contact:

Re: 4-way intersections: Throughput and deadlocks [image heavy]

To test it, you can throw it on the test bench. Download my version of it if you want to understand the test bench more easily.

It doesn't make any unnecessary crossings, no roundabouts and is correctly signaled. It should perform exactly like the compact intersection or really any good 4 way 2 lane unbuffered intersection.
There are four unnecessary chain signals in the center.

Is you want more throughput you have you to add buffers in some way.
Updated version of aaarghas intersection testbench. viewtopic.php?f=194&t=46855&p=383830#p383830
My Highscore intersections: https://pastebin.com/dGsJT71Z
My 3 to 1 train merger https://pastebin.com/AcLHXcwd

Requia
Long Handed Inserter
Posts: 70
Joined: Thu Mar 24, 2016 12:24 am
Contact:

Re: 4-way intersections: Throughput and deadlocks [image heavy]

I widened the multicross interior so that I could fit a roboport in the center and I got slightly better performance. 89-->91 on the p1 test in hansjoachim's test bench.
https://pastebin.com/14CXYw5B

hansjoachim
Fast Inserter
Posts: 122
Joined: Wed Apr 26, 2017 7:03 pm
Contact:

Re: 4-way intersections: Throughput and deadlocks [image heavy]

Requia wrote:
Thu Apr 25, 2019 1:46 am
I widened the multicross interior so that I could fit a roboport in the center and I got slightly better performance. 89-->91 on the p1 test in hansjoachim's test bench.
https://pastebin.com/14CXYw5B
Great
91 is within the margin of error. As Aaargha wrote in the first post:

I let it run for 15 min while counting how many trains exit the intersection. When the time is up I calculate the average throughput in trains/min. Note that this is not too exact and small measured differences between intersections may not really mean that much.
Updated version of aaarghas intersection testbench. viewtopic.php?f=194&t=46855&p=383830#p383830
My Highscore intersections: https://pastebin.com/dGsJT71Z
My 3 to 1 train merger https://pastebin.com/AcLHXcwd

Tallinu
Fast Inserter
Posts: 119
Joined: Sun Jun 14, 2015 8:14 am
Contact:

Re: 4-way intersections: Throughput and deadlocks [image heavy]

Let's see... http://fbpviewer.trakos.pl/b/OoOTd5Vodk ... tHjTMrkK2w

Interesting adaptation. That 6x6 square in the center means it would still have room for LHD signal placement. Would only be able to put a medium power pole in there with the roboport, or a nearby substation could be used to power it...

I had originally gone for the closest possible track placement to minimize the time required for a train to make the crossing, but two track segments is undoubtedly not the biggest performance factor.

I've made a note to change that in the "official" versions next time I fiddle around with them. (Who knows when that will be though, heh)

yvanaquino
Burner Inserter
Posts: 5
Joined: Tue Feb 14, 2017 12:00 am
Contact:

Re: 4-way intersections: Throughput and deadlocks [image heavy]

Is it possible to submit a design for testing and improvement/feedback? I helped my kid develop this but it's his intellectual property.

hansjoachim
Fast Inserter
Posts: 122
Joined: Wed Apr 26, 2017 7:03 pm
Contact:

Re: 4-way intersections: Throughput and deadlocks [image heavy]

yvanaquino wrote:
Tue Apr 30, 2019 7:36 pm
Is it possible to submit a design for testing and improvement/feedback? I helped my kid develop this but it's his intellectual property.
Yes it is=)
But please provide a screenshot so that one doesn't have to open factorio to see it.
On this forum I'd prefer 4 way intersection. The trainstation looks decent, you could add some more signals so that a train is right behind the train at the station. Apart from that you could probably make it smaller, but I never cared for saving a bit space or rails.
Updated version of aaarghas intersection testbench. viewtopic.php?f=194&t=46855&p=383830#p383830
My Highscore intersections: https://pastebin.com/dGsJT71Z
My 3 to 1 train merger https://pastebin.com/AcLHXcwd

lee1026
Burner Inserter
Posts: 5
Joined: Thu Jun 22, 2017 11:10 pm
Contact:

Re: 4-way intersections: Throughput and deadlocks [image heavy]

Does today's patch (0.17.38) change all of the deadlock discussions?

hansjoachim
Fast Inserter
Posts: 122
Joined: Wed Apr 26, 2017 7:03 pm
Contact:

Re: 4-way intersections: Throughput and deadlocks [image heavy]

lee1026 wrote:
Fri May 10, 2019 4:29 pm
Does today's patch (0.17.38) change all of the deadlock discussions?
It probably does, needs to be tested first.
Updated version of aaarghas intersection testbench. viewtopic.php?f=194&t=46855&p=383830#p383830
My Highscore intersections: https://pastebin.com/dGsJT71Z
My 3 to 1 train merger https://pastebin.com/AcLHXcwd

SkiCarver
Inserter
Posts: 42
Joined: Sat Apr 06, 2019 9:36 am
Contact:

Re: 4-way intersections: Throughput and deadlocks [image heavy]

Similar to some of the designs posted, but it works well.

Left Side Drive

Right Side Drive
Attachments
train.png (1.89 MiB) Viewed 46 times

SkiCarver
Inserter
Posts: 42
Joined: Sat Apr 06, 2019 9:36 am
Contact:

Re: 4-way intersections: Throughput and deadlocks [image heavy]

4 lane version of the above junction.

Sorry to say I have not tested this, as I dont use 4 lane rail networks.

Attachments
4.png (3.16 MiB) Viewed 42 times

Who is online

Users browsing this forum: No registered users