A request for feedback from server operators

[FlowerChild]

I have a question: Does world size affect server performance?

I know Jadedcat believes it does, but I'm personally not convinced. Obviously the number of loaded chunks will affect performance, but to me, that should be independent of the size of the savegame, and a large save size should have a negligible impact on performance.

So I was wondering if other server ops could give me an idea of their experiences in this regard. It can have rather dramatic consequences on how I design the mod (I actually have one feature on my todo list right now that I should probably scrap if it's the case), so I want to make sure that I'm 100% clear on this one.

[BinoAl]

Most of the issues I've seen are actually related to paid hosting services: Some use smaller SSDs, and can have some issues if they get close to filling up their alotted space. For the most part, it only matters how many chunks are loaded

[FlowerChild]

That's what I personally would have expected. If you fill up the space at your disposal, you'd run into caching issues, much like if you start running out of hard-drive space locally.

If that's correct, then what we're basically looking at are server specific issues that can be addressed with more storage, and in that case, I won't adjust my design to accommodate it.

One of the big draws of MC is obviously the "infinite" world, so this whole thing about basically restricting player expansion to prevent running into performance problems struck me as rather odd.

Would love to hear from other server ops on their experiences related to this.

[morvelaira]

I have to say, I'm not sure the size of a world affects its performance, but I do know there is a certain bit of conception out there that this is the case. The only server I've been involved in that had a rather large map was not one I was an admin on, unfortunately. The one I'm running currently has been slowly, but steadily expanding, and while there's certainly performance detriments when expansion is underway, once that's calmed down I don't see anything worse for wear.

[FlowerChild]

but I do know there is a certain bit of conception out there that this is the case.

Hehe...yeah, but as you know, such popular beliefs are often entirely inaccurate. If server-space filling up causes caching issues, and if many servers have rather limited space at their disposal, then that would certainly be enough to create such a belief amongst gen-pop.

[Ozziie]

I have the most basic level of paid server hosting and have had issues using several mods with an average number of performance hogging entities.

As Morv also mentions, I notice a significant performance hit as the server/world is generating new chunks.

On the same server running BTW I now have the largest world I've ever had on the server and as of yet have had very few performance issues. The amount of entities I am using is at least equal and if not more than previous.

I don't wish to speculate too much but I would imagine the number of chunks loaded at one time by different people would affect server performance, but this is not the same as generally having a large world. I imagine entities within these loaded chunks would compound the problem and may lead to the assumption that large worlds = impact on server performance.

[jakerman999]

I've done a bit of testing in preperation for setting up a server for a friend, although it's been automated and not performed under real world scenarios.

Under the most common circumstances(reasonably sized SSD, fair amount of RAM) larger map file sizes have no detrimental effects to runtime speeds with up to 40 simultanous "players", up to a 8192 chunk radius.

Using an HDD instead, introduces a small delay, but never seems to reach a full second in lag time. It seems to be related to number of chunks loaded, as only having 5 "players" doesn't have the delay. It was a 7200RPM, haven't had a chance to do a 5400.

Using a RAMDSK provides the most intersting results. Regardless of number of "players" loaded, delay is effectively 0 beneath 256 MB world size. The delay increases by roughly half a second for every 256 MB threshold of world size. You could probably have up to a gig before playes started complaining about speeds, but I think the delay could be lowered by having faster RAM or system clock times, or the threshold might be increased by having larger RAM chips installed.

*Players in this experiment consisted of stripped down linux clients inside VMs running minecraft and an autohotkey script that made Steve move in a circle.

[DaveYanakov]

The only performance based reason we try to keep the world saves on our servers small is to keep the regular backups speedy. If too many chunks are generated the process starts to get a bit lengthy. Other than that and disk space concerns, the only performance hits I have ever seen could be blamed on how many active players there are or the server host itself.

Saves can get up into multiple gigs very quickly if you have people scattering to the winds on a well populated server but the objective based gameplay we have seen to date serves to keep that in check. No one will want to build a remote base fifty kilometers out and then have to spend eight hours walking back to spawn when the time comes.

Edit- I did forget one potential downside to large worlds. People traveling on a rail network are causing chunks to load at obscene rates. When you have settlements spread out and the current nether glitches cause people to get stuck in limbo, you get very large rail networks. One trip can demand that a server load a few hundred chunks every minute.

[jadedcat]

I would like to point out the specific world in question, the one where I believe world size was a problem was 62,528 blocks by 42,619 blocks and over 4 GB in size. Its kind of an extreme case as on the other servers I have played and done admin work on I have never seen a server over 15k squared.

[SterlingRed]

Before I closed my server my world size was roughly 1gb. The physical size of the world made no difference in performance from day one to the end. I considered limiting the map for two reasons. 1) To encourage player interaction by making them closer to each other rather than 8k blocks apart. 2) To keep storage space under control. I was keeping as many as 7 backups each taken four hours apart so that I could roll back in case of a cheater significantly affecting the world without making my players loose days of gameplay.

You have fixed issue 1). So the only reason I can see now is some admins may need to limit maps due to storage limitations of backups. Admins who are more able to actively manage the server than I was wouldn't need as many backups.

[FlowerChild]

Yup, encouraging people interacting is good, but there are also certain resources in the game that you may need to explore quite far out to acquire (mycelium for example).

If map-size is an issue, that means I need to not design features that require exploration, especially if those resources are finite and players will need to move ever further out to find them as a server is going for longer and longer.

Frankly, I'd rather not feel obliged to avoid doing that.

[PatriotBob]

I've been hosting small MC/BTW servers on local machines for a year or two now. Generally on terribly old, single core, DDR hardware. So anything that affects performance tends to show up in a bad way fairly quickly because the hardware is already under significant load anyways.

One particular map we expanded to have some 10km area squared minimally and I can say that chunk generation aside, it had no performance impact. We could play just as well regardless of how much terrain was generated.

Side note to server admins: If you're keeping worlds small to make backups easy, you're probably doing it wrong. Take a look at keeping your world on a ZFS store. You'll be amazed. (You'll have to host on Linux and use the user-space driver or dedicate another machine for storage) But look, I run things on 7 year old hardware, so this isn't about throwing money at a problem but more about using the best tools for the job.
\distracting rant

[duartemad]

The vanilla server I used to play would have borders on the worlds to minimize the risk of map corruption

[Marasambala]

I reallly don't think world size is the issue, we're using large biomes and on top of that the biome distribution hasn't exactly been even. We had to cover a lot of ground to find them all. Some folks are still exploring just because that's what they like to do.

I dare say the BTW tech does make a difference, though. I've found it really doesn't matter how many players are on, the server load is about the same. The RAM usage has steadily increased as we establish more automation. At one point we had a large concentration of BTW tech just through our main nether portal, and the server was exceeding the RAM alotment, and crashing, more often than not when anyone passed through into the nether. We managed to get the stuff removed, and we haven't had a problem since. We are slowly and steadily approaching my max RAM and I'm going to upsize this week.

[FaceFoiled]

I don't think world size causes any issues related to BTW. I can only speak for Edolas though, as I have not played on any other BTW servers. One thing to consider though, is that this was a pretty big map before adding BTW, and I know there have been corrupted chunks in it as well (pre-btw). But here are a few of my thoughts on the matter;

Lag in railroads;
Traveling between places has increased a lot with BTW, thanks to the additional hardcore modes and, and general difficulty that people strive for in their automation designs. As mentioned before, traveling over a long distance with a rail network, can load chunks at pretty crazy rate. If I am not mistaken though, this was already a bit of an issue on basic Minecraft with long railroads on straight lines (render distance in tunnels).

Server restarts (manual restarts);
As far as I can tell, this is the only thing that has changed since BTW was added. The memory usage does increase over time, even when very few players are online. Server restarts are the only way to resolve this, which was not the case pre-BTW. Again though, I doubt that this is related to the size of the world itself. A bit of brainstorming went on at a thread or ingame chat about this, and one thought did stood out as "plausible". This concerned the continued processing of resources, but not the gathering of those. For example, mills that keep grinding netherrack, but not being collected (stacking of entities). Not sure if that is even possible in unloaded chunks though.

[embirrim]

I can confirm what FF said, specially in the nether where we have our base a long way away, the server always increases memory and cpu usage quite a lot when travelling.

I started hosting my server on a small laptop, and I can say that I had to restart the server after every play session. As soon as somebody joined for the first time after a restart, even if they left, server memory usage would start to increase. Server restarts are really needed imo, the game becomes unplayable.

As for map size, our map isn't that big atm but exploring hasn't increased average memory usage. I did use to have a 1.2.5 map, SSP, that had enormous decreases in performance after I went looking for mycelium. No idea if that's related or helpful.

[BigShinyToys]

I can confirm what FF said, specially in the nether where we have our base a long way away, the server always increases memory and cpu usage quite a lot when travelling.

I started hosting my server on a small laptop, and I can say that I had to restart the server after every play session. As soon as somebody joined for the first time after a restart, even if they left, server memory usage would start to increase. Server restarts are really needed imo, the game becomes unplayable.

As for map size, our map isn't that big atm but exploring hasn't increased average memory usage. I did use to have a 1.2.5 map, SSP, that had enormous decreases in performance after I went looking for mycelium. No idea if that's related or helpful.

This sounds more like another memory leak in minecraft than something BTW related.

[FlowerChild]

I don't think world size causes any issues related to BTW. I can only speak for Edolas though, as I have not played on any other BTW servers. One thing to consider though, is that this was a pretty big map before adding BTW, and I know there have been corrupted chunks in it as well (pre-btw). But here are a few of my thoughts on the matter;

Face, would you mind running a test for me?

I'll put an option in the next release to disable my reliable block-update code. If you could try turning it off on your server and see if it resolves the memory leak issue, I'd be very much obliged.

It would at least help me rule out that one as a potential culprit, as it's the only thing I can think of in the mod that could potentially cause such a problem.

[FaceFoiled]

Sure, not a problem at all. :)

[FlowerChild]

Sure, not a problem at all. :)

K...excellent. Added to my todo list :)

[andreinawolf]

Hopefully this is the right place to post this :) I found this thread after searching for "reliable" trying to find a discussion about the "fcEnableReliableBlockUpdates" option that was added in 4.22.

I started doing quite a bit of server jar modding to help with the problems lifewater's server has been encountering. Our tick time was going very very slow (which ended up being about 460 villagers in a 1x2x1 space at spawn) and occasionally the server would seem to crash when using a portal to the nether. There were also complaints about very few mobs in the world.

After dealing with the 460 villagers I put in a lot of time measurement code. I found that two things were happening:

1) chunks were staying loaded where there were no players. these chunks were increasing the tick time, but more importantly they were increasing the overall area that mobs could spawn in, making it seem like there were barely any mobs in areas where there were players.

2) the number of entities tracked by EntityTracker for the Nether and End can get very large over a somewhat short amount of time. it doesn't always happen, but when it does the number of entities can be as high as 70,000 overnight. usually it's all pigmen, but sometimes it will be a small amount of pigmen and the rest will be XP orbs. this only happens when someone has visited the nether or the end at least once since the server had been started. when someone enters the nether or the end when it's in this state, the server takes a while to remove all the entities that have built up, and the server seems to crash. eventually it recovers on its own and you can connect again.

Foolishly, I waited till earlier today to start playing around with fcEnableReliableBlockUpdates. Disabling the the reliable updates seems to fix at least problem 2, but I haven't had enough time to let the server run on its own to see what will happen.

I heard from someone earlier tonight that this was also happening on "It takes a village", so I wanted to share this on the forums.

I hope this information helps!

[Marasambala]

I only just got around to trying the server config option, and immediately saw a very clear difference when I changed the option in the config file. I've been expressing frustration with my internet connection on my server and had concerns that I would no longer be able to play SMP. When I changed the config option, my lag problem immediately went away. I restarted right before I changed the option and then again after I changed the option. For me, BTW SMP went from unplayable to playable, again. Looking back, I believe my problem coincided with the 4.20 update.


edit: For what its worth, I wasn't having any problems in either single player or LAN.

[LaserSushi]

I find world size effected my old server, but alas my old server was crap (really just a outdated PC set up to host as a linux based server). So long as you are running on a modern machine and making sure you don't add too many players it seems fine. It seems like the more chunks that contain players the worse the lag gets.

Anyhow thats just my experiances.