Belt rebalancer is not your friend

[User_Name]

All-right, after doing some quick tests for the Faster belt corners thread I switched to other save and built proper test facility.

We have 3 very long and 100% compressed belts.
As soon as I power up the system, inserters start collect copper as fast as it arrives. Throughput is limited by the bottlenecks.
Central belt has no bottleneck, thus it have 100% throughput.

I cut off the power when end of copper on the middle line reaches finish line.

Test 1: Straight belt vs Split & Rejoin vs Fast corners



Straight: 1260
split & rejoin: 1264
Fast corners: 1241

Results are the same as in previous test, just more accurate.
Split & rejoin is as fast as the straight belt - no loss of compression at all.
Fast corners lose 2 - 3%, which is acceptable for the most cases.

Now to the interesting part:
Straight belt vs Slow corners vs Belt rebalancer.



3..2..1..GO!

Visual representation:


Results:

Straight: 1312
Slow corners: 869 = 34% throughput loss
Belt Rebalancer: 840 = 36% throughput loss

So... belt rebalancer is not your friend.

Save file: https://dl.dropboxusercontent.com/u/343 ... test_2.zip

[Piranha]

Can you do the bottom test again but with the belt rebalancer as 2 splitters therefore cutting out the U-bend on the one side.. I doubt much will change but I always wondered.. maybe make it a 4th lane so you can compare it separately..

Can't find a great picture for it but the upper most 2 splitters on the left side is what I am describing.. middle lane is the feed and the upper 2 splitters would still go 50/50 then attach back onto the middle but with on 1/4 turns instead of a full 180 on one side.



Also I don't know if it makes a difference but I usually would leave 1 space after the top splitter in the middle so it was a 50/50 feed instead of 25/50/25 which I guess doesn't really matter and is probably better the 25/50/25 way.


Anyways, thanks for the post. I always like reading these informative posts about how to optimize my factory and its lines.

[User_Name]

Can you do the bottom test again but with the belt rebalancer as 2 splitters therefore cutting out the U-bend on the one side.. I doubt much will change but I always wondered.. maybe make it a 4th lane so you can compare it separately..

Tested 2 more rebalancers:

1) "Overkill"


1 side:


2 sides:


rebalancing pattern


2) "2 Splitters"

Rebalancing pattern:



1 side:

2 sides:



Rebalancing pattern slightly differs, no throughput loss for both designs.

[User_Name]

And finally...

2 splitters, separate (the spacing in between is not necessary)

Rebalancing pattern:



1 side:



2 sides:



Rebalancing pattern is as good as of common rebalancer, but no throughput loss at all.

[User_Name]

Well, rebalancing pattern is not so pretty if things are on the other side of the belt



So, the place for the perfect rebalancer is still vacant

[User_Name]

I nominate this ugly bastard:



(when you absolutely have to balance your belt evenly)

It has 100% throughput


Variant:

more

It has same 100% throughput as his ugly brother above, but more aesthetically pleasing. Uses two extra splitters that are not really needed though. The price of beauty.

We can add even more pointless splitters to this!

pic

[n9103]

Uh... rebalancers are always going suck when they use the same speed belts as what they're processing from and are restricted to a limited size.
Also, of the rebalancers you tested, only the traditional one and the last one evenly rebalance. The rest favor one side or the other.

Finally, the balancer you recommended before the last one isn't a balancer, but a divider. Works to shift a load that is only on one side to be evenly on both sides, but that's all it can do. Mixed or compacted inputs render the setup next to pointless.

Glad you added the last one though, as it's the only rebalancer that's still effective on express belts.

[User_Name]

Uh... rebalancers are always going suck when they use the same speed belts as what they're processing from and are restricted to a limited size.

Well, traditional one sucks even if it's in full blue on yellow belt. It loses less throughput, but still can't keep up with the straight yellow belt.
The last one doesn't suck if it's yellow on yellow. It's only downside is size.

[User_Name]

Can you do the bottom test again but with the belt rebalancer as 2 splitters therefore cutting out the U-bend on the one side.. I doubt much will change but I always wondered.. maybe make it a 4th lane so you can compare it separately..

Can't find a great picture for it but the upper most 2 splitters on the left side is what I am describing.. middle lane is the feed and the upper 2 splitters would still go 50/50 then attach back onto the middle but with on 1/4 turns instead of a full 180 on one side.

...

Also I don't know if it makes a difference but I usually would leave 1 space after the top splitter in the middle so it was a 50/50 feed instead of 25/50/25 which I guess doesn't really matter and is probably better the 25/50/25 way.


Anyways, thanks for the post. I always like reading these informative posts about how to optimize my factory and its lines.

Like this?



Unfortunately, still sucks.

[Marbles]

It is not a proper rebalancer, you need a piece of faster track where the two split yellow belts come together to ensure compression and throughput.

[User_Name]

It is not a proper rebalancer, you need a piece of faster track where the two split yellow belts come together to ensure compression and throughput.

It helps, but the problem is still there

Red:


Blue:


Full blue:

[Kanddak]

What about two splitters, with a gap preserving the rebalancing pattern, with faster belts on the remerge?

[User_Name]

What about two splitters, with a gap preserving the rebalancing pattern, with faster belts on the remerge?

I tested it before. Slightly better than the common rebalancer with red parts (because there is no U-turn), but still not quite there.

The fundamental problem is that it's impossible to insert items on the belt using T-joint and maintain 100% compression, even if it's yellow belt->blue rebalancer->yellow belt.
To get around this we can split the input in two, lowering the throughput requirements by the factor of 2, than use the double-sided T-joint on each half, and join them back into one, as demonstrated on the previous page

pic

This is proper rebalancer. There is no other way to keep 100% throughput.

[ssilk]

The question for me is: When do you really need a rebalancer?

I saw some lets play, where the player made a balancer at nearly every possible place. That is not needed. And in some cases it is contra-productive.

So let's look into this more detailed: When is a rebalancer needed?
Precondition: We have or can do on both lances of a belt the same item-type.

A) Inserters lay items only on far lane.

Example: Best example is a row of furnaces. One side is iron in, the other side is iron plates out. But then they are always on the far side of the belt. I need to move the items to the other side.

Answer: Not really always.

Throughput on the production-side: Can all furnaces output their furnaced items? That's normally not more possible after 7 or 8 furnaces (basic furnace). If you are below that: Where is the problem? You can't produce more. One lane is more than enough. If you rebalance that below this number of items, you generate a bigger storage on the belts. In fact you double the storage. But will that be faster?

This is a fail, if you produce expensive items (For example processing units) and limit production. The items on the belts cannot be used! You generate an unuseable storage.

B) To balance out that the inserters prefer to take from the inner lane

Example: You have some iron wheel production and the inserters empty the inner lane very fast, but not the outer. It happens, that one lane gets really empty (or near empty), but the other (far side) is still filled.

Answer: Here again is the throughput important. Some calculation: Are the iron wheel assemblies able to use much more than the capacity of one half belt? If not: Why changing it? The belt delivers enough material. Empty lane mean: there is no hidden storage and we have still capacity free.

If you are still unhappy with this: you have for example two rows of furnaces, one for the right and one for the left lane, which you join then with a splitter. Now this looks like so, that one row of the furnaces works fully, but the other side not. And the inner stacks of the furnaces are now full. So you loose capacity.

A balancer doesn't make sense, because it will change the picture so, that the first furnaces (both sides) are working, but the last not.

I recommend then, to put the overproduction - and it is nothing else - into a storage (What I mean can be found in the wiki) to "stock-produce" iron-plates... And a good storage will balance the belts out by design.

This makes sense especially in the early to middle game.

C) Balance transport from source to destination to optimize throughput

Example: You have some kind of outpost and as it happens, most miners put the items on one side, so that one lane is full and some miners cannot output items.

Answer: Yes. Do it. But in this case you need normally only "one-side balancers" - a splitter and one belt.


TL;DR: Balancers are not needed, if you are looking for throughput. There is normally a misplaning in the factory structure.

[User_Name]

C) Balance transport from source to destination to optimize throughput

Example: You have some kind of outpost and as it happens, most miners put the items on one side, so that one lane is full and some miners cannot output items.

Answer: Yes. Do it. But in this case you need normally only "one-side balancers" - a splitter and one belt.

It will not give you full throughput at all times.

Let's say you have all your miners on left side (50% of the total belt throughput), but your demand is not constant, it jumps from 100% belt throughput to 0% every 10 seconds, which is 50% averaged, so you should be OK, but you need fully compressed belt in the end to satisfy 100% demand periods.

With only one side of the belt 10 sec moving, 10 sec idling you using only averaged 25% of total belt throughput.
You put a splitter plus 1 belt thingy to let the ore fill the (empty) right side.
At this moment you expecting to have 50% averaged throughput from your miners to your consumers, right?

Well, not so fast. When there is a demand, you have 50% throughput, but when the left line is clogged (not moving), you have only ~30% of the throughput on the right line, because T-joint can't compress items well enough.

Proper rebalancer does not have this problem. It handles "1 clogged 1 empty" input at the maximum speed.

Of course, this is edge-case scenario, but you have this problem pretty much always when you try to rebalance partially clogged belts with splitter+belt thingy to increase total throughput.
It's just not so obvious.

[User_Name]

There are two cases where rebalancer may be a good fit:
1) to get maximum throughput from unevenly clogged belts
2) when you putting things on the busy belt with inserters, and want to maximize output

[ssilk]

Of course, this is edge-case scenario, but you have this problem pretty much always when you try to rebalance partially clogged belts with splitter+belt thingy to increase total throughput.

Well, yes, of course.

I wanted to say throughput is not always needed. Sometimes it's more clever to be thrifty: does anybody need a double line of green inserters? Of course not.

So we speak here only from copper and iron ore or copper and iron plates. And a handful more, which needs this.

And my thesis is, that in most cases this isn't needed, if you balance that during building. Simplest example: place miners equally on both sides. Or built furnaces (and others) always symmetrical. That spares you a lot of complicated belt handling. (Of course this is sometimes fun)

It's just not so obvious.

Totally right. This is complex, but good design will become accepted.

[roy7]

This thread is brilliant. I just discovered it. Thank you for doing these tests and providing all of the examples and images.

[dee-]

This is the rebalancer I use for my standard "two-lanes-of-fast-belts"-system.

a.png (715.22 KiB) Viewed 14036 times

It is able to rebalance every single one of the incoming four lanes equally onto the four outgoing lanes:

image

b.png (543.37 KiB) Viewed 14036 times

If one or more of the four outgoing lanes are blocked the other not blocked lanes still get filled with more items.

It also preserves the full item throughput if no rebalancing is needed:

image

c.png (865.16 KiB) Viewed 14036 times

Red corners are needed to keep the full thoughput of the yellow lanes.
(I think the red corners in the center merging area might be downgraded to yellow as the throughput there is only 1/2 yellow belt. I haven't tested it yet because it just works However keep in mind 180° u-turns behave differently when they side-feed onto a belt)

Red belts on the field where left and right belts merge are needed to have the side-merges working. Yellow belts would be too slow to get all incoming items on the merged belt.

The rebalancer also only needs equal or lower technology of your usual belt system.


I'd be delighted if you could put it though your testbed
It should be possible to split it into a left and right half if you only need single-belt width.

[Trepidati0n]

Props to you. What I have noticed is that a double splitter balancer takes up no more space but gets you close to perfect. Very low cost with massive improvement. Thanks!