Page 17 of 36
Re: 4-way intersectios: Throughput and deadlocks [image heavy]
Posted: Sat Nov 11, 2017 10:01 am
by aaargha
Divaya wrote:Sure, I'll work on the design a little more. I definitely like the idea of allowing multiple trains to make left turns in the intersection.
Edit: Not as hard as I imagined it'd be! Here's the second iteration of the intersection, and it allows opposing trains to make left turns at the same time.
*images*
I tried a similar approach on the 4.5 guage rail distances, but I had trouble with tight areas around the left turns that made it unfeasible to signal.
Added as Traditional with some slight signalling changes, I'd advise you to be careful with the double rail signals on exits, as they're so close trains may stop in the intersection if the output backs up. Throughput is very reasonable considering its size, it does become about five times wider, and taller, with the lane changers I'm using though
Zijkhal wrote:So, I've done a bit of testing with the Pasta because I was dissatisfied with the 66 / 73 throughput values, and found that it can actually do up to 83 / 85. The problem is the exit of the intersection. With leaving a train's length after the exit, trains follow very far behind each other, limiting the throughput. With using the one signal every two cars method, the throughput improved to 80 in set 1. At this point I realized that the higher throughput I was consistently measuring was due to me putting signals as close to each other as possible on the exit line, so trains were following each other closer.
Anyway, I was thinking, does the 1 train long rail block has to be right after the exit merge of the intersection? As trains accelerate, there could easily be a gap longer than one train between them, and it could be used to unlimit the throughput while maintaining the behaviour that trains only exit the intersection if they can fully leave the exit merge, thus not blocking the entry, and not degrading the rating to B.
So, I have created this:
https://pastebin.com/Dmq0Xvyh
In three trains length it ensures that trains only exit the intersection if they can fully leave it, but it does not limit the throughput of the exit line.
Aargha, could you please retest the Pasta with this merge thing attached?
While this kind of optimization can be powerful it also limits the intersection to one train size. If we use different train sizes we may end up in situations like below where trains can park in the intersection if an output backs up.
- mixed.png (112.82 KiB) Viewed 10447 times
I will not allow this kind of optimization as I feel that A) fixing train size is too limiting, and B) I'd have to redo all tests for the comparisons to be fair.
However if you're OK with fixing trains to one length then you can also modify this optimization to work on the internal buffers of the intersection. If trains are a fixed size you may also disregard the train-length output block in some situations if the distance between intersections is a multiple of the train length. I've been meaning to make a post on these kinds of optimizations for a while now but I keep getting distracted by other projects
Re: 4-way intersections: Throughput and deadlocks [image heavy]
Posted: Sat Nov 11, 2017 11:06 am
by Tallinu
Zijkhal wrote:Anyway, I was thinking, does the 1 train long rail block has to be right after the exit merge of the intersection?
Merges do not require a full-length exit block. As you've found, they perform better if the first few blocks are quite short, on the order of 3-7 track tiles. Block length can scale up with distance from the merge as train speeds will have increased and the gap between two trains will be wider (but any very high traffic area will typically perform better with blocks no longer than 7 tiles, or
maybe 14 if you expect consistently high speeds -- exactly two or four car-lengths, respectively).
Unfortunately if a merge is too close to an intersection you can't do this without risking deadlocks. And a merge containing a full-train-length block results in decidedly sub-optimal throughput.
This is why the junctions I've posted since finding this thread (with rare exceptions) include enough buffer space after
all intersections so that all trains can stop if necessary and wait for their turn to merge without risk of blocking intersections. It makes the minimum size of the junction longer, but it also has the benefit that you don't need to provide additional anti-deadlock buffer space between one such junction and
any other junction
of any type! The built-in buffer spaces do that job already!
Zijkhal wrote:So, I have created this:
If you're adding on something that long, you'd be better off just adding that one train-length of buffer space after the last intersection and moving the merge to the end of that buffer. It'd be shorter than that bluerpint and wouldn't have any reliance on train length.
(Some slight digression: Merges could perform significantly better if train stopping / speed control logic included a little bit of prediction regarding other trains moving through blocks the train needs to enter, checking its speed and position to predict when the block will become clear, but who knows if that will ever happen. The only existing way I know of to improve merge throughput is with circuit-controlled signals, and it's far from perfect.)
Re: 4-way intersectios: Throughput and deadlocks [image heavy]
Posted: Sat Nov 11, 2017 4:09 pm
by Zijkhal
aaargha wrote:
I will not allow this kind of optimization as I feel that A) fixing train size is too limiting, and B) I'd have to redo all tests for the comparisons to be fair.
Fair enough.
Well, then, here are my "fixes" for that:
The Mk II:
https://pastebin.com/qFmsfJRW
Same size, but has more than one crossing on the left turning lanes after a buffer.
The Mk III:
https://pastebin.com/VpZk3Yh6
Larger, if I am not mistaken, this is the only intersection that includes a buffer before and after every crossing. I am not sure about the throughput gains over the above one, however...
Re: 4-way intersectios: Throughput and deadlocks [image heavy]
Posted: Sat Nov 11, 2017 11:01 pm
by Divaya
aaargha wrote:
Added as Traditional with some slight signalling changes, I'd advise you to be careful with the double rail signals on exits, as they're so close trains may stop in the intersection if the output backs up. Throughput is very reasonable considering its size, it does become about five times wider, and taller, with the lane changers I'm using though
What do the lane changers you're using look like?
Re: 4-way intersectios: Throughput and deadlocks [image heavy]
Posted: Sun Nov 12, 2017 10:56 am
by aaargha
Divaya wrote:What do the lane changers you're using look like?
I'm using the ones Tallinu designed a
while back, I've added that to the OP.
Zijkhal wrote:aaargha wrote:
I will not allow this kind of optimization as I feel that A) fixing train size is too limiting, and B) I'd have to redo all tests for the comparisons to be fair.
Fair enough.
Well, then, here are my "fixes" for that:
The Mk II:
https://pastebin.com/qFmsfJRW
Same size, but has more than one crossing on the left turning lanes after a buffer.
The Mk III:
https://pastebin.com/VpZk3Yh6
Larger, if I am not mistaken, this is the only intersection that includes a buffer before and after every crossing. I am not sure about the throughput gains over the above one, however...
And oh boy, did those fixes pay off, both the MKII and the MKIII are among the top designs so far, great job!
The only thing I did was to lengthen a few of the output buffers that were a few pixels too short:
- too-short.png (182.71 KiB) Viewed 10409 times
In other news, the OP is now about two thirds of the post limit, I'm both anticipating and dreading us getting there
Re: 4-way intersectios: Throughput and deadlocks [image heavy]
Posted: Sun Nov 12, 2017 11:57 am
by Zijkhal
aaargha wrote:
And oh boy, did those fixes pay off, both the MKII and the MKIII are among the top designs so far, great job!
Yes! My life is complete
aaargha wrote:
The only thing I did was to lengthen a few of the output buffers that were a few pixels too short:
too-short.png
Crap. Noobie mistake.
but I'm wondering, does lengthening those buffers so they can fit ~7 cars make sense, so the average crossing speed of the 6 car trains will be faster?
Re: 4-way intersections: Throughput and deadlocks [image heavy]
Posted: Sun Nov 12, 2017 9:51 pm
by aaargha
It might, the only way to be sure is to measure it. I might set up some small test to try to do that after I've finished my current project:
If I still have my sanity when I'm done I'll take a look at some simple buffer length tests.
Re: 4-way intersections: Throughput and deadlocks [image heavy]
Posted: Mon Nov 13, 2017 12:39 pm
by ElderAxe
Can you please test this one?
Re: 4-way intersectios: Throughput and deadlocks [image heavy]
Posted: Mon Nov 13, 2017 9:44 pm
by Tallinu
Zijkhal wrote:aaargha wrote:
And oh boy, did those fixes pay off, both the MKII and the MKIII are among the top designs so far, great job!
Yes! My life is complete
aaargha wrote:
The only thing I did was to lengthen a few of the output buffers that were a few pixels too short:
too-short.png
Crap. Noobie mistake.
but I'm wondering, does lengthening those buffers so they can fit ~7 cars make sense, so the average crossing speed of the 6 car trains will be faster?
aaargha wrote:It might, the only way to be sure is to measure it. I might set up some small test to try to do that after I've finished my current project:
If I still have my sanity when I'm done I'll take a look at some simple buffer length tests.
I've done tests along those lines and posted the results a few pages back. On mobile so hard to go find them, but some extra space definitely does make intersections available sooner. Depending on train length and braking speed. The test trains can benefit from up to around 7 track tiles of extra stopping space if I remember right.
Re: 4-way intersections: Throughput and deadlocks [image heavy]
Posted: Thu Nov 16, 2017 11:05 am
by aaargha
DarkElder wrote:Can you please test this one?
Added as "Almost squareabout", the unresolvable deadlock is a 4-way right-to-straight path change on the outer lanes.
Tallinu wrote:I've done tests along those lines and posted the results a few pages back. On mobile so hard to go find them, but some extra space definitely does make intersections available sooner. Depending on train length and braking speed. The test trains can benefit from up to around 7 track tiles of extra stopping space if I remember right.
Oh right, I'd forgotten about that, sorry. I believe
this is the post.
Re: 4-way intersections: Throughput and deadlocks [image heavy]
Posted: Thu Nov 16, 2017 5:41 pm
by Tallinu
Yep, that's the one. Because the effect is greatest for the first few pieces of added track, even a small amount of extra space can make a big difference, but there is that minimum crossing time to consider. Now all of those tests were done using the testbed map which has full braking force researched and the trains were all running on rocket fuel, so there may be worse performance in real game situations if either of those things are not true -- lower braking force would increase braking distance, requiring more extra track to get the most benefit, and running on coal or solid fuel might reduce acceleration and make minimum crossing time higher as well. I hadn't thought to run those tests using other fuel types, but it should be easy enough, I had included a blueprint for the measurement rig I was using.
Re: 4-way intersections: Throughput and deadlocks [image heavy]
Posted: Sat Nov 18, 2017 12:48 pm
by ridesdragons
you made a castle lol. I actually like that design. I'm gonna have to show some of my friends lol
Re: 4-way intersections: Throughput and deadlocks [image heavy]
Posted: Sun Nov 19, 2017 1:47 pm
by ElderAxe
aaargha wrote:DarkElder wrote:Can you please test this one?
Added as "Almost squareabout", the unresolvable deadlock is a 4-way right-to-straight path change on the outer lanes.
There shouldn't be a deadlock on the outer square. It's not a big square roundabout. trains can only go straight using that outer lanes.
Also i made an OCD friendly version of the same design. That would be great if you add this one too.
Re: 4-way intersections: Throughput and deadlocks [image heavy]
Posted: Wed Nov 22, 2017 10:47 am
by mrvn
DarkElder wrote:Can you please test this one?
Why isn't it symetrical? That usualy means you get one speed in one orientation and another when you rotate it.
Re: 4-way intersections: Throughput and deadlocks [image heavy]
Posted: Fri Nov 24, 2017 10:42 am
by ElderAxe
mrvn wrote:DarkElder wrote:Can you please test this one?
Why isn't it symetrical? That usualy means you get one speed in one orientation and another when you rotate it.
The one on my previous post is symetrical. OCD Friendly Version
Re: 4-way intersections: Throughput and deadlocks [image heavy]
Posted: Sat Nov 25, 2017 1:18 am
by Jap2.0
The "I don't know what to call this - Mk. 1"
This is somewhat of a roundabout while somewhat not. Trains can only go straight and right, but can turn around after the intersection, so a right turn can either be straight, then right or right, then straight. I have no idea if this will deadlock or how well it will do.
Re: 4-way intersections: Throughput and deadlocks [image heavy]
Posted: Sat Nov 25, 2017 8:36 am
by Zijkhal
Jap2.0 wrote:The "I don't know what to call this - Mk. 1"
This is somewhat of a roundabout while somewhat not. Trains can only go straight and right, but can turn around after the intersection, so a right turn can either be straight, then right or right, then straight. I have no idea if this will deadlock or how well it will do.
I think you should try widening the straight going tracks at the entry / exit, so the turnaround tracks dont have to cross it, which could improve throughput. Also, I'd merge all three lanes as close to each other as I can, and use chain signals for all three instead of merging two and then a chain signal when merging with the straight one.
Re: 4-way intersections: Throughput and deadlocks [image heavy]
Posted: Sat Nov 25, 2017 3:55 pm
by Jap2.0
Zijkhal wrote:Jap2.0 wrote:The "I don't know what to call this - Mk. 1"
This is somewhat of a roundabout while somewhat not. Trains can only go straight and right, but can turn around after the intersection, so a right turn can either be straight, then right or right, then straight. I have no idea if this will deadlock or how well it will do.
I think you should try widening the straight going tracks at the entry / exit, so the turnaround tracks dont have to cross it, which could improve throughput. Also, I'd merge all three lanes as close to each other as I can, and use chain signals for all three instead of merging two and then a chain signal when merging with the straight one.
Okay, I hadn't thought of doing the turnaround in the middle. As for the merging, some of the spacing was a bit weird, so fixing that would definately help. I'll keep all that in mind for mk. 2. Thanks!
Re: 4-way intersections: Throughput and deadlocks [image heavy]
Posted: Sun Nov 26, 2017 8:55 am
by gremblor
I really like Tallinu's multicross design, but I run a 4-track 3-8-3 (14 car) rail network, and his blueprints for this train length are only available in 2-track form... I built a 4-lane RHD multicross scaled up to 14 car buffers, available here:
https://factorioprints.com/view/-KzrTRj_P9IqOK9K8ryQ
Re: 4-way intersections: Throughput and deadlocks [image heavy]
Posted: Mon Nov 27, 2017 12:04 am
by Tallinu
gremblor wrote:I really like Tallinu's multicross design, but I run a 4-track 3-8-3 (14 car) rail network, and his blueprints for this train length are only available in 2-track form... I built a 4-lane RHD multicross scaled up to 14 car buffers, available here:
https://factorioprints.com/view/-KzrTRj_P9IqOK9K8ryQ
Actually, you can find the original, slightly older, 4 lane version of the Multicross in the first post of the thread, right at the top of the 4-lane "A" section.
There's only the LHD version of what I'd now call a partially "extended" layout, which contains a couple of extra buffer zones that aren't really worth the extra space with longer trains, though.
I had intended to polish up some rescaled blueprints for both RHD and LHD (of the non-extended version, since there's so little point to the extension in anything but a throughput test environment) and add them to my drive folder with the others... But I've just been too busy and stressed with life and other things, unfortunately.
I'm pleased to hear that you like the design enough to do that much work on it yourself though.
