Page 1 of 1

Testing maximum numbers of players

Posted: Thu Apr 21, 2016 13:22
by Mogoloko
Hi all,

My wife is a secondary school teacher here in Spain. A few days ago, she went to a conference, and a speaker spoke about Minecraft EE. She's very interested in use it in her classes. When she told me, i suggested to try some open source alternative, and we discovered Minetest. I'm web developer and gamer too, so i'm also very excited with the idea.

In a few hours we had some basic but great ideas, but we needed technical support about minetest servers. Basically, we need to know how many users (more or less) can play in a server over LAN. And external players connections (server 16GB, 10Mb/s up, 10Mb/s down)?

So we decided to search in the net these question, but we've not found anything.

Wouldn't be interesting if the community organize different tests in different servers, and publish the results?

For example:

Server: wwww.myserver.net
Computer: i5 8GBRam
Spped: 10/100 Mb/s
Max-players: 20
Result: No lag


Thanks

Re: Testing maximum numbers of players

Posted: Thu Apr 21, 2016 14:02
by KCoombes
You would also need to standardize the mod installations, or track only vanilla minetest game servers, as the mods can and do add to the lag.

Re: Testing maximum numbers of players

Posted: Thu Apr 21, 2016 14:30
by Mogoloko
Yes, i'd be fine.

Anyway, even if each server which make the test use differents mods, the feedback would be very interesting, and would help to get an idea of what you can reach.

Re: Testing maximum numbers of players

Posted: Sun Apr 24, 2016 05:27
by DI3HARD139
I've hosted upwards of 75+ players over 7 servers (2 with over 270 mods and the rest lightly modded (20ish mods)) and used around 4-6Mb/s of a 13Mb/s connection. The amount of mods loaded onto the server will change the bandwidth requirements. Most bandwidth consumption will come from a client that hasn't connected to the server before. Mods that use the most bandwidth and/or system resources are mapgen mods, mobs, technic, mesecons and pipeworks. Check my signature for my server specs.

Re: Testing maximum numbers of players

Posted: Sun Apr 24, 2016 19:37
by Mogoloko
Thanks DI3HARD! Your reply gives me hope!
We'll starts some tests in a few weeks. We'll host a wifi lan game, without mods, in a classroom (30 students).
The first thing we want to know is if we can succesfully play Minetest with 30 players but in a wifi network with over 200 students connected.
Second step would be the same test but with two classrooms (60 students).
Finally, repeat the same testsbut with external connections (no lan).

I'll post the results here.

Thanks a lot!

Re: Testing maximum numbers of players

Posted: Tue Apr 26, 2016 21:43
by Errol
Please keep us posted, Mogoloko. I'm also interested in how heavy a load the servers can take. :)

My server is not on a lan, but on Digital Ocean. I'm curious to see how many people I can host, and 75+ makes me hopeful as well.

Re: Testing maximum numbers of players

Posted: Wed Apr 27, 2016 07:26
by Mogoloko
I'll do, Errol. We have another member in the project: another teacher (History of art) from the same high school. So we'll be able to test with 2 classrooms over the same server.
We'll start to do the first test in early June.

Re: Testing maximum numbers of players

Posted: Thu May 12, 2016 18:26
by iZacZip
It depends on how much space your RAM/Memory disk can hold.
I've been on server that can only hold round 12 players,
On the other hand, one that can can hold 64+...
So, really - there's no exact point that a server can hold.

Re: Testing maximum numbers of players

Posted: Mon Jan 11, 2021 18:06
by AllanWindmill
Any updates?

Re: Testing maximum numbers of players

Posted: Sun Jan 17, 2021 03:54
by Minix
This is really interesting because it is about taking the minetest engine to the limit, I think minecraft servers running on modern hardware (i9 10th gen) can handle around 200 players. Since minetest is more efficient, it could possibly take more players than that using the same hardware, it would be really awesome to see a powerful server CPU, like an Epyc AMD server CPU running a minetest server. From my experience hosting servers, for 500 players in a default MTG survival server with no additional mods we would need:

-10 MB/s minimum network bandwidth and up to 35 MB/s for peaks for uplink and 5 MB/s and 17,5 MB/s for downlink respectively
-Really fast storage with good random read/write performance, like NVME SSDs
-RAM is going to be an issue if you have a big world, so when players start exploring it takes more RAM, since this depends a lot on player behaviour you could get away with just 4 GB of RAM if players don't move around very far but for that amount of players I think 128 GB of RAM is a good amount
-The CPU is complicated because it also depends on player activity, if a lot of players generate new mapblocks fast (let's say sailing with boats) the CPU will be hit hard, but if players are just walking around the same place the usage will be very low. We also have to remember that while mapgen can take advantage of multiple threads, everything else runs on just one thread. So for CPU we would need something that has very high single core performance and can be overclocked to further improve performance.
Now, for players generating new mapblocks with boats all at once we would need the equivalent of 250 Pentium 4 HT 530 CPUs, so I believe a Ryzen 3xxx or Intel i9 10th gen can easily handle that (don't quote me on that though)

This is just me rambling, I didn't do much math for this

Re: Testing maximum numbers of players

Posted: Sun Jan 17, 2021 04:16
by Festus1965
Minix wrote:
Sun Jan 17, 2021 03:54
This is just me rambling, I didn't do much math for this
The used mods are very important, as they can take the server down by too much work for ABM, technique, automation, animation and mods ...
AS of the limit of older this server-list gamer counting without loos point in rank,
I am sure that some of the good hardware-servers can handle 100 with a low profile mods usage.
But most of this servers are rented, shared and with slow swap memory ... what might low down there max.

Might set up a just real MTG server without any further plus mod, and test it ... I think a Intel 4 single core, might be able to handle it, as most work is storage (hdd > ssd > NVM) read, if not kept most even in Memory (32 GB) using keep cache and a good internet connection for deliver.
Without more mods it might be no problem for 4 core CPU, near 3 GHz, as I only see them used a bit when mapgen is forced.

Re: Testing maximum numbers of players

Posted: Sun Jan 17, 2021 04:24
by sorcerykid
Minix wrote:
Sun Jan 17, 2021 03:54
-Really fast storage with good random read/write performance, like NVME SSDs
For any serious Minetest server, RAMdisk is the way to go. I use an 18GB tmpfs partition for my worlds directory, backed up to physical media daily with rsync. It makes a significant difference in terms of performance, esp. where random access is concerned.

Re: Testing maximum numbers of players

Posted: Sun Jan 17, 2021 20:53
by The32bitguy
I should mention that there is huge potential to optimize web traffic on servers by compressing textures and models for all of the mods. Thanks to exe_virus#4413

https://github.com/ExeVirus/Compress-Obj
so max allowable compression of mtg objs: 71.5% of original size
Absolute Best PNG compressor: https://github.com/fhanau/Efficient-Com ... l/releases
It was able to compress all of default's png textures from 243,330 Bytes to 138,388 Bytes. And those have already been mostly optimized. I even wrote a dead simple windows batch script that auto compresses every png in the current folder and subfolders. I.e. you can just drop this into your server /mods folder (back it up of course) and compress everything in one go.

Re: Testing maximum numbers of players

Posted: Mon Jan 18, 2021 04:51
by Minix
Thanks 32bitguy, I compiled ECT and gave it a try, I'm going to use it for optimizing my server's media.