I get it, EVE is old and the back end cannot be easily optimized without large changes in the source code. From what I’ve heard every system uses only one thread (=1 CPU core) on the server to handle the simulation in that entire system. Multi threading is a bitch to get working and keep in sync, I know from experience. It’s not financially feasible to pay developers to make large changes in the core of the back end which could take months to develop, test and release, nor is feasible to upgrade entire hardware for tens of thousands of dollars every year just to keep a few large battles from lagging out.
Why not buy a few “heavy load” boxes that can at least handle these large battles a little better, for a fraction of a cost of a regular server? I know that this will be a blasphemy to all enterprise system administrators, but hear me out.
Get a handful of regular boxes with binned i7 8700K, overclock them to >= 5.2GHz, cool them with water and assign these boxes dynamically to high load systems (CCP seems to be able to do that already if given heads up about a large fight coming up). These i7 have a much higher single thread performance than CCPs current Xeon E5-2667 v3 that are used as nodes (Cinebench single thread performance - 8700K is 30% faster with stock clocks, probably 40%+ faster when overclocked) and the i7 are tens times cheaper than Xeons.
Sure, it won’t be ideal, it is not a fix to the problem, but merely a temporary improvement. Even with i7, things will still slow down, but a performance improvement of close to 40% for a fraction of a cost of a regular enterprise level server is not that high of a risk? I know that companies and system administrators think it’s a unfathomable to use consumer chips in a production level enterprise network, but these boxes wouldn’t be used 24/7, so their reliability risks diminish heavily. There’s a reason why some of the largest hosting companies in the world like OVH use 7700K for game servers in their datacenters.
Obviously I’m not CCP server engineer, I don’t know how their network, virtualization, SANs, load balancing, database clusters and all that are set up, but I think that a performance increase of 30-40% during battles for a low cost (compared to the other available solutions) would be easily noticeable and appreciated by a lot of players, corporations and alliances.
The first easy step to improve those mega battles would be to make drones, fighters and maybe even missiles sub-par for those events. My sniper ship worked OK yesterday. Could warp to the keepstar (got in with the Jackdaws), orbit and shot. Guns were cicleing and damage numbers showed up.
I believe that the dedicated nodes CCP uses for places such as Jita and those systems that have been flagged by the “Fleet Fight notification” are already very high spec (prototype maybe even?) machines.
Large fleet fights are at the limit of the technology (hardware wise).
Multi-threading is a possible answer but that would mean a ground up re-write, which is not something CCP can afford or (I doubt) has the technical staff for these days.
Ah, thanks for the link. I see now that they have what they call “Everest nodes” which is exactly what I suggested here. I guess my thread is pointless then, the only way to improve from current situation would be to refactor the back end, which is a huge task with large risk for little benefit.
Even if they improved the performance, more players would file in to lag it up again.
CCP knows this. They redesigned sov to split up large fights to avoid exactly this kind of thing. They just didn’t have the foresight to understand that these large, spammable structures would be centralized points of conflict important to the sov game. They designed around themselves.
They know that those mega battles draw attention to the game
They know that there is a hardware limit how much entities the servers can handle per system/grid
They know that drones, fighters and missiles increase server load and acted accordingly in the past (Grouped missiles create one strong instead of several regular ones, no more than 5 drones and fighter squadrons act as a singular unit)
They started to spread fights over several systems with “fozzie” sov.
And then we got a lot of stuff that does quite the opposite of those points. Like we saw it yesterday in 9-4RP2.
As we don’t see such big server crushing battles for the smaller citadels one idea could be like this:
Spread the fight for the large ones a bit over the system? Large Citadels have X (2?) Power generators that have to be anchored one lightyear away in the same system. You have to shoot/defend both, the citadel and the generators. Same with extra-large citadels, only that the two generators have to be in the same constellation. Same rule, you can only attack them all at once, no use in going for a single generator at a time. The generators don’t get destroyed until the citadel does, they only deactivate under a certain amount of DPS. With deactivated generators the large and extra-large citadels have lower resists or something like that.
And shift the meta a bit to make sure Dreadnaughts are the best option to bring to such a citadel timer fight. Or even make gun boats with no drone bay the most ISK efficient way to fight in large scale battles in general. Maybe even give them a small damage bonus for grouped weapons. Some drawbacks in speed, warp speed, agility, application or whatever could assure that they don’t get used for every role in the game and don’t make other ships types obsolete. No new ships, just a tweak in stats of the ones we have.
Because I think we could have far less TiDi in huge fights when the server had only to manage one group of stacked guns for most of the ships involved.