Player Features and Ideas Discussion

You do realise that we're working with a server shard, not one server, right?

Long story short, eve runs on a huge cluster of servers. Normally, each server has multiple solar systems running on it, one python process pr solar system (I believe), because there's no reason to dedicate one server to each solar system. This is, however, done when we're expecting large fleet fights (or in some cases, this reinforcement request is never sent), so that the fleet fights have as many resources available to them as possible, to limit lag.

Unless I'm mistaken, everything within a system, from the market, stations, POSes, PI, pilots, drones, missiles etc etc etc, is processed within that one solar system process. The rest of the eve universe isn't anywhere near that process. This is why, when the solar system isn't reinforced, and a huge fight goes down, sometimes you see solar systems at the other end of the universe starts bitching about lag.

Add to all this the fact that lowsec lags even harder than nullsec because of extra rules, then add in LOS calculations ... and what do you think will be the result?

If you absolutely wanted to make LOS etc work, you would have to drastically rework how eve handles players and solar systems etc. If you make each player a python process, then the load would be much more evenly spread out across the nodes, and things like LOD would be slightly more feasible. You'd still end up with issues, chief amongst would be concurrency issues, which I believe is the main reason why we're still at the point where everything that goes on in a solar system is in one process.

As for offloading the calculations to a separate server or core, that's just adding complexity to an already complex setup, it's yet another thing that can go wrong. If there's two things you'll learn if you setup larger systems for a living, it's "keep it simple, stupid", and "what can **** up, will". It can work, yes, what if that (or those) servers are loaded to the point of taking more than a second to do the calculations? What if it dies?

I'm not saying it wouldn't be hilarious, it would. It would be awesome to actually fly the ship as well, but there are some things which just aren't possible, either realistically, or without making huge changes to the existing system architecture, and them's the breaks.

The risk as it's structured now is we have essentially 2 servers (well, that's oversimplifying it a little, there are proxy servers etcetcetc in the mix as well, but if they don't work, nothing works), the solar system server and the SQL server (which is a cluster). That's two servers which can have issues, be it hardware, software, external hardware (network, power, etc).

If they moved the LOS calculations out to a separate server, that's 3 servers which can have an issue. And that's ... more. The nature of what's beinc calculated is irrelevant, except if you're thinking there could be bugs, in which case I'll just point out that the more sourcecode there is, the more bugs there are. And what happens if the calculations server goes down? The solar system server is now dependent on the calculations server doing its job.

Additionally, this is dependent on the server being able to work properly concurrently, i.e. it can fire off the calculation and keep doing all the other computation that needs to be done. If it can't, all you've done is split some functionality out to a different server, you'll still have to wait more or less the same amount of time for the calculations to be done (maybe even more, since you have to collect, serialize and deserialize the data, do the calculations, serialize and return the data, deserialize the data and apply damage etc to the correct ships/modules/etc). We're talking about more that has to be done pr tick, and I'm not convinced the gains would be worth the extra risk and costs, not with the current architecture of the entire server cluster.

In short, it isn't "far less risky", it's additional risk. Each server you add to a solution adds risk. You have harddisks that die, memory that go bad, motherboards that fry themselves, network cards that stop working, network switches that stop working, servers losing power, etc etc etc. The more you have that can go wrong, the more will go down.

I'm an absolute noob in programming but...

Is it possible to make each grid a separate process, instead of every system ?

This would remove the concurrency problem while still increasing process granularity.

I think the tool Veritas use to break down how much time is being spent in which functions showed that a fleet warp was taking something like 100-150ms to initiate one fleetwarp. I'm not sure those numbers are accurate (it's been a while). You can do the math from there.

The actual "formations" isn't the problem, it's the LOS that is the problem.


If you've seen how much of a ***** it is to move a player from one solar system to another, then you wouldn't ask that question.


IIRC, a grid is generated every time you land somewhere.

ok, forget the line of sight thing, how about formations just for having your squad in the right places so everyone is together and it looks cool,

Aye a easy first step would be to implement formations and it would look real cool if you had 2 enemy fleets flying in formation opposite to each other firing!



I doubt it's possible to make a 100 pilots fly in a perfect formation at the exact same speed..