Discussion: long-range logistics revamp


  • Dev Team

    @atanner This would be an inter-room structure. With many of them, it would create high concurrency pressure on rooms synchronization tick stage and affect ticks performance significantly.



  • @Atanner's micro-terminal idea is interesting. It's harder to implement though. On the surface it would be another global intent which is bad.

    With some tradeoffs terminal transfers can be done via 2 room intents, one to decrease the sending terminal's resources and a separate one to increase the receiving terminal's resources. To prevent a resource race condition (possibly duplicating resources) the terminal would need to lock out withdrawals for a tick or two.

    The locked out withdrawals could be disguised as "reserving" the micro-terminal before sending. Or it could be disguised as lag, the sent resource are immediately deducted but don't show up at the destination until the next tick.


    Kinda nerd sniped myself with how to to interlock a terminal send using only room intents.

    On tick 1:

    • The sending terminal issues a buffer intent with resourceType and amount.
    • The processor records the resourceType and amount it deducted from the terminal's store in a _buffer property.
    • The Terminal tick handler saves any _buffer property as buffer.

    On tick 2:

    • The API sees the buffer property and automatically issues a blank receive intent for the destination terminal.
    • When the receive intent is sanitized or saved the values from the buffer are attached to the intent (prevent intent hacking).
    • The processor increases the store of the destination terminal (truncating to fit).
    • The terminal tick handler deletes the buffer property.

    This is would work for regular terminal transfers. I haven't thought about how it would interact with the market yet. But aside from possible market interactions this would allow players to go back to a 0 cooldown terminal and dramatically decrease the amount done in the global tick processing.



  • @artch This would also make harvest boost slightly more useful as it would decrease the amount of warp container decay while mining if I understand correctly.



  • @deft-code

    @deft-code said in Discussion: long-range logistics revamp:

    I like both ideas of putting a minimum range from sources and decaying the containers from a harvest action. @U-238's idea of allowing the warp-containers to be closer but increasing cost seems like a good one (more remote mining options the better). I think the decay should be based on energy harvested from the action rather than the action/intent itself otherwise supersizing the harvesters could bypass much of the problem.

    Even cooler if we get a new visualization in the client to "show" the sources reacting with the warp-containers when a source is harvested.

    What if there is no minimum range for the structure itself, but the decay is dependant on the distance to the .harvest action, sort of like how towers are more effective on short ranges. This could allow WC directly adjacent to the source, but at a high decay cost, scaling down with the range to the source. Make it super expensive when close to the source, then lower costs based on longer distances.


  • Dev Team

    The root post has been updated. Let's perform the second iteration of the discussion!

    The primary change is that we split warp containers maintenance cost onto a regular repair (mostly like usual containers except warp containers will survive much much longer) and activation cost. Also, .harvest operations will increase decaying speed so now warp containers should take their place as one of the alternative ways of transferring energy from remotes instead of creating the single best solution.

    👌


  • @o4kapuk said in Discussion: long-range logistics revamp:

    Math.min(0, 11 - range)

    I assume that's meant to be max?

    I guess it's also per-intent? I feel this cost is likely to be lower than it first appears in practise. We already see oversized harvesters to save CPU. Right now for me on MMO my code thinks the sweet spot is around 14 WORK, though that number goes up and down depending on factors. Call it 15 WORK, and now I'm paying a third of the cost than I would be if I had 5 WORK. From 5 energy/tick at 1 range to 1.666 energy/tick. Couple that with the fact I only need charged containers for 1/3 of the ticks, and things are getting pretty cheap agian.


  • Dev Team

    @tigga

    I assume that's meant to be max?

    Yes, fixed, thank you.

    About oversized harvesters (which obviously costs more in energy and spawn time), do you have any numbers to share, or was it just from the top of head?



  • I think if you give me 1 CPU I can get ~13 energy profit (as of right now, it varies). This is based on predicted income - cost and adjusted for measured income - cost. The difference is usually pretty small.

    Put in a lot of sums and that works out as ~12 WORK/creep for maximum profit. 7 extra work costs me ~0.6 energy/tick and saves me ~0.12 CPU/tick, so you can see it kinda works out. The marginal cost on adding an extra 2xWORK, 1xMOVE is ~0.17 energy/tick, ~0.012 CPU/tick saved, so not worth doing at 13 energy/CPU (but would be at 15 energy/CPU). This assumes spawn time is not a consideration which is generally true either on shard 3, or at high GCL.

    Assuming this code doesn't have bugs (obviously not a certainty) I think it's reasonable to expect people to have at least 10 work/harvester in remote rooms, halving peak decay to 2.5 energy/tick. Going to 15 work/harvester would cost ~0.4 energy/tick/harvester, and save ~0.8 energy/tick on the warp container, so I'm pretty sure that given I'm already nearly there that's a no-brainer. Going to 20 starts to hit diminishing returns (0.4 energy/tick for larger harvester, ~0.4 saved on the warp container) so stopping at 20 is probably reasonable.

    Now with 20 harvest/tick I don't actually need the warp network online all the time. Exactly how the "online" method works isn't clear. Seems like 1 energy/tick while online for the whole room. I probably only need it online 1/3 of the time as I can clear a source in under 100 ticks, saving ~0.67 energy/tick.

    So we're getting to the stage where a 3 warp container room with spread sources will cost 1.5 energy/tick (container maintainance) + 0.33 energy/tick (activation cost) + 2.5 energy/tick (harvest intent decay) + 3.6 energy/tick (harvesters). That comes to ~8 energy/tick and ~0.15 CPU/tick for two sources throw in a reserver, road maintainance and other random intents and maybe we can call it 10 energy/tick and 0.2 CPU/tick. This blows my current energy/CPU out of the water - I can get ~4x the energy per CPU spent if I use 3 warp containers/room naively.



  • Is warp-container charging a creep intent or will they self change from energy stored?

    What happens to the warp-container's contents when they are not charged? Do the resources drop? Or are they just inaccessible (like a perma-cooldowna) until charged?

    Do each of the warp-containers need to charged with 100 energy every 100 ticks or just one of the warp containers?



  • Deft, I expect that there is literally only one of them, with one storage. Just with 2 or 3 .pos objects that you can interact from. Since a creep can issue only one intent, it's safe from 'accidentally' accessing both pos points, and the contents can only be withdrawn the next tick.

    Oh! Oh! Or there are 2 or 3, but they simply share the same store object. Yeah, that's better. The others are just dummies for the UI. and pos point.


  • Dev Team

    @Tigga thanks, your calculations do match mine primarily. How much you spend on haulers currently?

    Now, in your assumptions, do you count transfer intent? Even for 15+ WORK harvesters (harvests in 1/3 of ticks) if could worth 0.9 energy/tick per source. We can activate warp containers by 300 ticks for 300 energy, so saving on activation won't be possible.

    @deft-code

    Is warp-container charging a creep intent or will they self change from energy stored?

    They are self-charging from their store.

    What happens to the warp-container's contents when they are not charged? Do the resources drop?

    They drop onto the ground. We can't keep them inside because this way it's possible to fill them with minerals (either due to user's bad code or attacker's malicious code) efficiently preventing further recharges.

    Do each of the warp-containers need to charged with 100 energy every 100 ticks or just one of the warp containers?

    It's 100 energy/100 ticks per room, not per warp container.



  • @o4kapuk said in Discussion: long-range logistics revamp:

    @Tigga thanks, your calculations do match mine primarily. How much you spend on haulers currently?

    320 energy/tick for 156 sources right now.

    Now, in your assumptions, do you count transfer intent? Even for 15+ WORK harvesters (harvests in 1/3 of ticks) if could worth 0.9 energy/tick per source.

    Yeah, I did. With 20W+8C you're doing 0.05 CPU/tick for harvest and 0.025 CPU/tick for transfer, which is where the 0.15 comes from.

    We can activate warp containers by 300 ticks for 300 energy, so saving on activation won't be possible.

    Sounds good. With all these things there's a balance between giving opportunities for smart code and forcing people down a given path I guess. Syncing stuff to reduce this upkeep would be smart code, but also maybe the "only way". Doesn't make a huge difference to my above estimates anyway.

    Really the "issue" seems to be that the majority of CPU in remoting goes into haulers - about 65% in my case. Remove this and you require 3x less CPU. Tying upkeep to harvest intents just leads to fat harvesters (which aren't a bad idea anyway for CPU).

    I'd also like to point out I'm doing the naive calculations. It's possible I want to put the warp containers further from the source and use high carry/low move haulers (eg. 20C/5M) to move from harvester to warp container. That'd be low CPU and might be cheaper than warp container decay.


  • Dev Team

    This post is deleted!


  • @o4kapuk said in Discussion: long-range logistics revamp:

    They drop onto the ground. We can't keep them inside because this way it's possible to fill them with minerals (either due to user's bad code or attacker's malicious code) efficiently preventing further recharges.

    Dropping on ground seems very burdensome behavior. Would it be possible to make defunct warp containers have an energy reserved store of 100 units that creeps can put into a container to recharge it? That store represents what the container will use for it's next recharge cycle, and gets automatically refilled from general store if there is 100 energy available.

    Then dead warp containers have their main store locked until they are recharged, but the resources don't fall on the ground. I'm not even sure what fall on the ground means given that the stuff inside isn't inside any one container.

    Random other thought: What if you could use batteries with warp containers for energy supply? Just as a possibility for actual things a commodity can be used for.


  • Dev Team

    @Davaned We thought about it. But that sounds like a great complication of the whole unified Store API or a weird exception out of it creating unobvious edge cases. Dropping resources seems more like in the spirit of these structures (aimed to serve as transport, not storage)



  • Yeah I can understand that, although this is a case of an energy-powered structure so having a .energy specific store is pretty intuitive, imo moreso than "one type of the resources you put in the general store disappear". How will the dropped resources behave then? Randomly sprayed under each warp container? This will drastically increase the pain of having skipped ticks.

    Also, withdrawing all energy from a container before it renews becomes extremely powerful as harassment. If they had a 100 .energy buffer that keeps gets refilled internally/from creeps that can't be withdrawn from it would take 100ticks of room control to have the network fail vs 1 tick.

    Formal suggestion: Internal deposit-only .energy store (like a nuker) of 100. This gets consumed every 100 ticks to maintain the network. If it's empty it gets refilled as highest priority from the internal store, but can also be creep deposited. Reduces complexity of warp container usage too since its more fault-tolerant and you have security that the container will be powered for next round.



  • This just seems like one of those ideas where it feels unnecessary. Why not flesh out power creeps more or make commodities less useless. More depth there would help more than another min/max tool that a small percentage of your playerbase will use

    👍


  • @o4kapuk Do you have a rough ETA for the terminal throughput limit by chance?


  • Dev Team

    @tehfiend It's soon to be published at the PTR

    👍


  • I know I'm too late to have my input considered with this update, but: Reading though what everyone else says here, warp containers sound like a "best solution" to remote mining (less cpu and energy cost/tick). The fact that they use energy, and decay costing energy seems overly complicated to upkeep (vs links with use energy and containers that decay).

    I'd rather see structures that don't require creeps to preform tasks on them for them to do their job (like links, labs, terminals). Warp containers still require creeps to move their energy across room borders, and increase code complication but less cpu. If something like pipelines, that could pump less energy the bigger the system is, were introduced, I would think you'd see haulers and pipes depending on the situation (removing the best solution problem).

    With respect to the terminal change, I'm not sure why civilians have to see the consequences. Just that same that structures can't be destroyed when hostile creeps are in your room, limit/block terminal calls.

    An alternate suggestion to blocking terminals: Make mortars a thing. Essentially mini nukes that can be loaded up with different resources (where certain resources have different effects, just as boosts help your creeps, these will take away from enemy creeps). For instance if KH is loaded, it will reduce any creep/structure that has a .store property by some percentage.

    I personally would like to see more inter-room actions, instead these updates would lessen inter-room actions (limiting terminals) and strengthen intra-room actions (with warp containers). lets see more things like mortars and pipes.

    Also, I'd be interested in seeing what internal suggesting the dev team came up with even if they were considered too much of a change that they'd break peoples code.