Player Features and Ideas Discussion

(My own feauture/idea concerning your code for Drones is at end)

I've noticed that TIDI and lag are pretty extreme in this game. In my own gaming codes (WC3/Starcrafft 2 custom maps for instance) I've often managed to eliminate hoards of useless code lagging up the system, code that intuitively does NOT seem useless.

I'm going to give you two real-life examples from my own programming.

---

Example 1:
"If Object A and Object B are more than K distance apart then..."

"distance" is calculated conventionally in 3 dimensions by the formula SQRT{ (x2 - x1)^2 + (y2 - y1)^2 + (z2 - z1)^2}

The problem here is the Square Root function, which runs an analytic (calculus-based) Newton–Raphson algorithm. Not only does this algorithm burn tons of computing power, but it's also UNNECESSARY.

We don't need to know the actual distance between these objects by executing the SQRT function, we only need to compare the sum of the squares against K^2. The boolean value (truth value) of the condition can be ascertained (for exponential less computing power!) by this code instead:

If K^2 > (x2 - x1)^2 + (y2 - y1)^2 + (z2 - z1)^2 is true, then [insert purpose of code].

All distances between objects on grid should be saved in their unresolved form (x2 - x1)^2 + (y2 - y1)^2 + (z2 - z1)^2 in the array holding those values. Even if an element of the array must be accessed to return the actual distance, just write your own function that estimates the square root to a single decimal place, this alone will save tons of more computing power! I very much doubt that EvE Online requires an accuracy 10-50 decimal places.

---

Example 2: Is the Distance Array full of redundant values?

There are n = 5 ships on grid: A, B, C, D , E

Is your array storing the distances between these ships a square matrix (the most inefficient matrix imaginable for this task?)

x A B C D E
A d1 d2 d3 d4 d5

B d6 d7 d8 d9 d10

C d11 12 13 d14 d15

D d16 17 18 19 d20

E d21 22 23 24 d25

Aside from the obvious that the distance from any ship to itself is zero (redundant and useless information), take note that the distance from A to C, d3, is also equal to d11, the distance from C to A. In fact, more than half of these values are redundant.

The above matrix requires n^2 calculations (5^2 = 25 calculations in this case), whereas only 4 + 3 + 2 + 1 = 10 calculations are needed, which is triangular number in the form of (n^2 + n)/2, for (n-1) it's (n^2 - n)/2 calculations. Therefore the number of calculations per game tick has been reduced by more than a half.

---

The moral of the story here is that any time there is a native function that uses a partial taylor series (trig functions for instance) or analytic function (pretty much any function that yields an irrational number) CHECK to ensure whether or not you actually need that irrational number output from the function to determine the truth value of the condition.

Make sure you don't have redundant entries in your array. A struct referencing a series of linear arrays of decreasing size is better suited than a square array in the above example. Check for redundancies!


I felt the need to write this thread because it is GLARINGLY obvious that your code (on grid) needs some serious efficiency overhauls. Simply using the lowest (tolerable) floating point alone can have tremendous results.

---

My own feauture/idea concerning your code for Drones: Drone Clouds.

Suppose there are N drones deployed by a ship. Instead of treating them as 5 individual objects, they should be treated as a single cloud. The Health/DPS of the cloud is scaled by N. Every time the health decreases by (1/N), the dps also decreases by (1/N) and a drone is permanently destroyed (N decreases by 1).

1: The center of the cloud is the reference point for determining distances and direction.

2: When the drones are idle about the master ship, the master ship is the reference point.

3: When the drones reach a target, if the speed of their target does not exceed their own speed, their position is ANCHORED to the target, and thus the target becomes the reference point for these drones.

4: If the target's speed suddenly exceeds the drone cloud's speed, the cloud unanchors regaining its own reference point.

5: If two or more drones clouds are anchored to an object (regardless of type of drones) they are adjoined to a Super-Cloud, greatly reducing the amount of reference points on grid.

6: Drone Clouds move at the speed of the slowest drone in the cloud --- this would be a balancing problem, but it's necessary to implement this idea. The drones need serious re balancing anyway, so when you get to them, this is a good time to include this change. It may also be possible to use the "average" speed and "average dps" of the drones. Consider this carefully if you elect to use the Drone Cloud idea.

Regarding performance, the physical simulation of the game is a relatively tiny portion of the game's processor load, thanks in part to quad trees and the fact that all physics in Eve is run with spheres. Go back and dig into some of the videos released around TiDi's inception along with Brain in a Box for what the main causes of load on the server are, but this isn't one of them.

Regarding Drones, drone clouds would have a major impact on the performance and balance of drones as a whole and as well as any flight that uses a mix of drone types.

The major skill element of drones is micro-managing individual drones that take aggro, which doesn't work with your cloud idea since if you're tracking individual drone HP and other characteristics then the performance increase of condensing drones into a cloud goes out the window. Plus the major source of DPS lost from drones is as a result of the activation of their microwarpdrives and the subsequent drop in applied damage due to tracking. This only works if individual drones are kept track of, since the range and falloff for drones is also small enough that those at the front of a chasing swarm have more chance to hit than those at the back.

Basically while this would likely result in a significant performance increase it also acts as a major buff to drones as a weapon system, while simultaneously removing the micro that characterizes them currently and acts as the only real element of skill involved in their use.

I'm all for checking for any code inefficiencies, always looking for ways to improve. For sure.

But,

-1 to the drone clouds idea.

Sure it may reduce load on servers, but it's at the expense of functionality and gameplay. As stated above, individual drone micro-management is a skill in itself, and being able to split drones to different targets is currently something that can be done. Maybe it rarely is, but it still can be, and removing that ability is a reduction in features.

Definitely -1

Perhaps it could be a system reserved for Super Nodes. The current drone system remains in effect until the fight is moved to a supernode.

A little icon (like jump fatigue, aggression icons) would signify this. There would also need to be a transition code between the node switch putting already deployed into their respective clouds.

I very much doubt the things you have mentioned would have any significance in a large sov battle.

heck, i didnt like them when they reduced the number of drones ships couldd eploy but "buffed the individual drones HP" cause im sorry, if im in pvp in lowsec and my drones are getting popped in 1 shot by a cruiser, they are still getting popped in 1-shot by a cruiser after their "HP upgrade" except now i have less of them out at once so i lose DPS quicker

but whatever, CCP will do what they want, and they prolly like the idea fo drone clouds,
*Snip* Please refrain from personal attacks. ISD Ezwal.

Lets see:

Optimization of code: I really hope they are already doing something like this, as it seems like just common sense, and since CCP is already pushing the edge of what hardware can be bought without exorbitant costs for prototype and pre-production hardware and we still have these issues, it really should have happened before now if it hasn't yet.

Drone cloud: Not so cool, because what happens when one drone dies and you have been kiting away from your cloud and then you drop another, and various other edge cases? That's part of the meta for ishtars right now, and they are used in fights large and small, and using 3 drones on deployment and then dropping 2 more to pull your range closer again would make them OP, similar to MJDs on the drone.

One shotted drones: Possible, but unlikely unless you have bad drone skills and unbonused drones, or they take the radical step of tackling your drones rather than you.

I have absolutely no comprehension of how to write code, but Saladinae makes it much more palatable, so thanks for that.

So, if each single drone has its own set of computer-lag-inducing issues but we can reduce it by making a cloud of them, but we want to be able to manage our lag-inducing drone pets independently because thats the way it is now, why not do a little of each.

For each drone of the same type launched from a single ship, they are considered by the server as one cloud.

If we look at the 500 ships on grid for a super-fight, and each of those ships launches 5 drones each, thats 2500 points of computing for the server (right?). If instead the server counts each ship's same-type drones as a single individual cloud, that drops it back to 500 points of computing. Even if there are multiple types of drones coming out of each ship, that could still be less than 2500 points (am I making sense?)

As far as lag goes, anything less than 2500 (for this example) is good.

So, for the individual, if I launch 5 same-type drones, the server tells me that my "drone swarm" (cloud sounds... weak) does [average drone dps] x 5 damage. When a single drone gets killed, its down to [average dps] x 4, etc, etc.

For the other guy shooting my drones: he targets one of the 5 drones on his overview, and begins to shoot. The server picks which of my 5 drones is getting shot at and once it pops it pops. The swarm is now down to 4 (or 3... or whatever...)

So, is it difficult coding/computing that a group of 5 drones from my ship is behaving like one "swarm," but still maintain the ability to scoop/shoot/disable a single drone?

Your interpretation is 100% correct. And yes, the server would random pick a drone to kill, you got that spot on. This also protects your drones from all being alphaed at once, as only one drone can die at a time, no matter how much damage the cloud takes!

And no, it's easy to scoop a single drone. The code looks like this:

Event: Master Ship selects "Scoop Drone 'X' "
Condition: If Drone Cloud is anchored

Action: Return drone 'X' to bay, decrease "drones deployed" by 1, and reevaluate vitals/stats on diminished drone cloud.

Since they are static, all information concerning them (distance to other points of reference) would only have to be resolved upon need. Objects static to the origin of the grid (origin is (x,y,z) = (0,0,0) ) are much easier to deal with in general. They wouldn't need a cloud.

That's not solving the original problem though, because you're either still tracking drones as individuals or you're assuming that anyone shooting that drone cloud is targeting the most damaged drone, which greatly simplifies targeting drones and focusing them down, which is not something drone users want.

Plus you need the ability to recall drones individually so you're firing off entity creation and destruction events as the clouds separate and merge, which are actually more expensive than maintaining entities in space, unless you're removing the ability to do this in which case that's an even bigger nerf to drones.

Plus if you're treating the drones in the swarm as individuals then you're tracking them as individual entities, they're just attached to a single in-space entity container, which eliminates a lot of your performance gains, since tracking and updating all of the individual entities is what actually drags down the server, not the drones actually being in space (especially since they're not physics enabled objects).