[SOLVED] best and cheapest hosting provider

I normally wouldn't make such a thread, but today I read a message on the Fresh Meat server that it will be shut down due to an excessive cost of $30 per month. This is not the first time that such news hit. Few people seem to know that:

A perfectly running 100% & 24/7 lag-free Xonotic server that supports 16 players costs no more $4.20 a month (often at special price only $1/month).

UPDATE: Right now it looks like $1 is totally overpriced, you get 4 dedicated Ampere/ARM cores with 24GB RAM on a 600Mbit connection in an undercrowded ARM-exclusive datacenter for $0 with Oracle Cloud Forever Free Tier (no hidden cost). That's the same pristine performance as 2-4 root servers. After extensive playtesting, so far everything checks out A+++ .

More details about the solution.

I will leave the rest of the thread as it was before the incredible $0 discovery.

WARNING: We don't know why, but after the free trial expires, some few people had all their servers deleted. Make sure that you never go over the limit, which at the moment is 4 Ampere cores and 24GB RAM total. This might perhaps happen as well if you run a test server with 10 cores just once, but delete it afterwards, within the free trial period. Always do rsync backups.

Another clunk we discovered is that sometimes it keeps the old "boot volumes", thus blocking you from creating new boot volumes for free. Simply delete the old boot volumes under "block storage" to fix this problem. This happens if you create e.g. 4 test servers, then delete the instances, then create 4 new ones. It might explain how people had their servers deleted after the free trial because they were actually overbooking and the new servers were all using storage that would have cost money.

---------- OLD ----------

It is therefore shameful to see long-loved servers with custom mods to shut down, because the server owners were sometimes paying 10x or 50x the actual cost with a better server provider for little to no actual benefit. With the time some of those servers have been running, they would have actually covered the cost for the next 15-100 years with a better provider.

For the last years, I believe the only server that actually *ever* exceeded 16 players at a time and for which the above price would not work out, has been eris/feris. This only proves the point that adding more beef to a server is pointless in almost all situations. I mean yes: maybe Fresh Meat actually one day had 20 players for half an hour or maybe it will have in the future and then with the $1 VPS 4 players will be disappointed because the server limit is only 16. But is it really worth it to pay $30/month instead, for just those very 30 minutes that happen only once or twice a year, and then whine about not being able to cover server cost and shutting it down all together? That is INSANE if you ask me. 

Of course I know that most cheap VPS providers suck. If you have tried cheap VPS you know the issue: there is poor load balancing between the customers (not so much a result of cheaping out the service, but rather a result of a shitty setup by the provider), and depending on the hour of the day or day or the week, you will get CPU chokes and lag spikes which can range from "annoying" to "unplayable". To some server admins I imagine, it is therefore a very seductive idea to just throw more and more money at the server hoster and fix the problem that way, which is probably the main reason why most server providers don't bother to fix their shitty setups on the low price tiers.

However, I have been in the IT business and OVH has been the provider of choice for a very long time. Historically speaking, it does not suffer from chokes, there are no lag skips, it just works and this has been so for almost a decade. They only get severe DDOS attacks once or twice a year which can cause lag for 1-2 hours, and sometimes their IPs are shitlisted because not just large businesses rent their servers there for $1000 a month, Joe Average also does. This means in short: the network quality is excellent and identical to a $50 root server.

OVH.com starter VPS is trial and tested for Xonotic, perfect with no drawbacks and many locations all around the world.






As you can see often and right now you get the special reduced price of $1/month for the exact same service, which makes it even more ridiculous and sad for people to pay $50/month instead. The special offer ends on 24.01.2023 14:00, which is 6 days from this posting. I think it occurs annually, or maybe even twice a year or maybe sometimes at random also, I am not exactly sure.


Now I have last tested OVH with Xonotic 0.8.2 in Frankfurt and France, I think a year ago, and I don't want to make any all-time guarantees here as Xonotic changes, the provider might change and it might also vary between locations. So before you start contracting for the next 5 years there, maybe at first use different payment method + credentials to get a test server for just 1 month at $1 or whatever.

But just today I have bought one year for 11 Euros (Frankfurt, Germany), the server is called "Extreme Voltage" and was configured from start to finish in less than 15 minutes via https://github.com/ballerburg9005/xonotic.us.to . We were only 8 players at most for testing, but judging from the CPU usage, 16 players is probably the minimum it will do, somewhere around 75-95% CPU usage. It runs of course butter-smooth as it always has been without any sort of skips or lags or CPU spikes.

I also made a second account under a different name and paid with Paypal instead of credit card to rent a second VPS for a year (Extreme Voltage Duel). I believe it is possible to use the same payment method multiple times with different accounts, but just to be safe, you could make multiple Paypal accounts as well.

I hope you also realize that it is obvously nuts to pay like 5x or 20x the price when the result will be identical 99-100% of the time, with the end result being server shutdown because of excessive cost. Of course if you use the server for other games or heavy stuff also, or if you are the one and only Eris/Feris, then it is not exactly the same matter. Plus you are of course free to personally pay whatever high prices you want where you want. But if it is about going bancrupt, begging or advertising about the money problem (no one in particular implied here) and having popular mods shut down then it should be a community concern and choices should be made foremost by economic criteria.


This is why I find it important that we can figure out a consensus here on which server provider is cheapest and best at a certain player count, because the variation between providers of what you get per dollar amount is absolutely HUMONGOUS. I believe it is very very difficult to contest the $1 VPS from OVH (update: funny how there actually was a $0 Ampere/ARM server with much more beef and even dedicated cores). And otherwise for other providers, it would be just a question of people reporting the numbers and doing the testing 24/7 for a whole week at least to see if the VPS will suffer from chokes and stutters (which most cheap VPS do).

It would be also be great if we could find a cheap provider that can handle more than 16 players (update: done! $0 Ampere/ARM = 32+ players). Even at OVH the twice as powerful server is quadruple the money (and 15x if you calculate bsed on the $1 special) ... so that's kind of a bad deal by comparison. Maybe we can get double the power for double the price (7x vs $1) somewhere else, that would really be a relief. But with time maybe some years down the road the $1 VPS will be twice as fast, probably, so it can only get better from there.

You are a brazen liar and advertise OVH here. Show me the output of vmstat -s on your $1 VPS.
This is a virtualization problem in general, I talked to the developers about this and found out how it works.
Often the negative effects of virtualization can be mitigated by a competent Linux setup. But it is impossible to get rid of problems completely, with 3 players and 20% CPU load on 99% VDS, periodic lags occur.
The virtualization problem is QOS(drop udp packets first) on the hypervisor, a large number of interrupts on multi-core hypervisors, stolen ticks, servicing people's backups for your money, rescheduling of processes by a scheduler on the hypervisor, the network interface buffer overflowed, etc.

The best thing I could find among virtualization is to look for vds on desktop systems with a 5Ghz processor. The fewer neighbors you have, the better your xonotic server will live.

The problem is that the server uses a fixed tick to calculate in-game 1/30s or 1/60s i.e. 16ms at 60 frames per second, if the server does not have time to do all the calculations in a tick, packets are lost.

You are a brazen liar and advertise OVH here.

Disagreement is ok and providing reason for disagreement is even much appreciated. But let's not insult other people, please. Thank you.

This is the output:

➜ ~ vmstat -s
1974224 K total memory
621408 K used memory
431684 K active memory
1092616 K inactive memory
214308 K free memory
47228 K buffer memory
1091280 K swap cache
0 K total swap
0 K used swap
0 K free swap
224073 non-nice user cpu ticks
8362 nice user cpu ticks
22699 system cpu ticks
5784365 idle cpu ticks
1425 IO-wait cpu ticks
0 IRQ cpu ticks
505 softirq cpu ticks
40322 stolen cpu ticks
877793 pages paged in
1818013 pages paged out
0 pages swapped in
0 pages swapped out
4027504 interrupts
6005104 CPU context switches
1674010265 boot time
27297 forks

(I was doing a lot of compiling today with the server, don't know if that matters. Currently I get like 2 stolen ticks per minute or so.)

It is of course true that the more VPS users you run on the same hardware, the worse each VPS will perform. However, the awful performance and "hits you in the face" lag experience you have with the average VPS provider is caused by poor/vanilla setups and bad load balancing. A small percentage of VPS users will max out disk IO for example, and this terrorizes all the other VPS customers because the scheduling is crap. The VPS then periodically freezes for 10 or 100 or 200ms at a time, which is extremely noticable and makes the game unplayable. On the other hand, if ticks or other resources are not stolen in huge bursts, even if they are stolen at the same rate, they don't cause noticable lag at all. Pair this with not having a ridiculous & unreasonable amount of VPS run on a single machine, and you got a totally lag-free VPS like OVH has. OVH is a huge and very established popular international provider. When you tell me that small or mid-sized VPS providers cannot figure out how to compete with their setups, because it just won't work like this out-of-the-box, this might very well be true. Like I said I have been in the IT business for some time, and it is just full of such night-and-day experiences all over, with services sometimes being 10x or 100x overpriced for absolutely no reason.

I don't particularly like to advertize for OVH here for free either. But keeping quiet about it and not being bold about it will only be harmful. So this should really be talked about and figured out by the whole community. Benchmarks and price comparisons should be public knowledge, or everyone is just bound to be fooled by a joungle of overpriced bad deals.

Everyone is free to play the server, see for yourself, rent one yourself for $1. Contest the performance with other providers.

It runs smooth without lag and stutters, about 16 players max, 24/7, all hours and days of the week, years ago and now still. Everyone can see that and I have yet to experience anything ever so slightly differently.

So if you call me a liar, as in not being able to believe your own eyes, I am even more so glad I made this thread.

Btw. someone told me in game that you can get some kind of unlimited/free Oracle cloud service, with 2 x86 vcores *plus* 2 ARM cores, or something like that. He sent me this video that I have not had the patience to watch:
https://www.youtube.com/watch?v=_m21Fxvu...wm-CtgFTb8

This looks to me like the one guy who said you can get $500 of cloud service credits for free each month, if you do some kind of tricks he didn't really explain. I tried to get into that, but it lead nowhere. Cloud services are a very bad idea generally speaking and I have no faith in them. I think it is just the usual scheme/scam. But maybe I could be proven wrong about this? That would be awesome.

40322 stolen cpu ticks

This is crap, not a server. That's a lot for a visited server!
The most tolerable is when you have less than 2k per day.
Moreover, you need to look at a loaded server, and not an empty one!
I played on your server and it's shit!

40322 stolen cpu ticks

This is crap, not a server. That's a lot for a visited server!
The most tolerable is when you have less than 2k per day.
Moreover, you need to look at a loaded server, and not an empty one!
I played on your server and it's shit!

You are right this seems to be a lot (like an impossible insane amount), but it totally contradicts other measurements and the general experience with the server, which is flawless for a VPS.

Since you asked about it the entire evening, the rate of stolen ticks has been a constant 1-2 per minute, and it is also only 2 under 100% CPU load, which I have just measured. This is very negligible and not noticable during gameplay, which is perfectly consistent with all the playtesting we have done so far.

Maybe what happened is that I have been throttled by the hypervisor from compiling so much code, downloading lots of maps and cloning git repositories today all at the same time. This could have caused this impossible amount of stolen ticks. The more unfortunate explanation would be that there has been a huge short-lived choke at a certain time of the day, because the service provider was overloaded, but I doubt this and have not observed this when I was doing in-game testing the entire day ... and it was not even peak hours ... plus it wouldn't really explain THAT many stolen ticks. I think what happened is that the VM was halted or severely throttled at night to do a full system disk backup image, which is automatically done nowadays on VPS. This is really the only plausible explanation for such a huge amount of stolen ticks on a system that worked fine all day long: a full system halt for 400 seconds at night.

So this is definitively something we have to keep an eye on. I will plot a graph from it so we can see at what hour of the day exactly ticks are stolen.

Since I didn't recall you amongst the larger group of players yesterday (where the experience was very positive for everyone but two people with bad Wifi and huge PL), and mostly no one was on it this afternoon when I was working on it, when have you even been playing the server? Did you join it for 10 seconds and then decided it must suck because it is so cheap? Enlighten us please what your experience was exactly, to arrive at this conclusion.

Dozens of people have played the $1 servers so far and the feedback was overwhelmingly positive.

We will see from the statistics I will be posting.

I realize there are no guarantees that the service won't degrade one day from too much users. But it has been perfectly fine last year and this year and I have yet to see a day where it actually sucked in any shape or form. Or else I would not have made this thread and speak with such certainty. I have done extensive testing already last year with Xonotic, and this year just seems to be no different. Years before that we were already using those VPS (but not really for real time stuff) and their performance was excellent as well.

Ok concerning the free Oracle Cloud thing I am trying it right now. First impressions:

It is quite beefy (more than twice the speed of $1 VPS), but it gets at least 1 stolen tick per second at 3am when idle. Also it just locked up in SSH for little to no reason. So I thought at first "surely this will suck too much". However then I played against bots for 5 minutes and it wasn't obvious at all that there could be anything wrong with it. I noticed two very very small skips, but this could have been my DSL line as well. It is called "FOREVER FREE ORACLE CLOUD SERVICE TEST".



We seriously need to reexamine our assumptions here. This graph shows stolen ticks per minute according to vmstat the last 5 minutes I played on the Oracle Cloud thing. 1 tick is 10ms as reported by vmstat, so with 3500 ticks stolen per minute that would be 35 of 60 seconds lost!! However the game ran smooth in actuality. So whatever is going on here with vmstat, it doesn't seem to make sense to say, that even totally ridiculous amounts of ticks that it reports as stolen will result in a bad player experience. Maybe someone knowledgable about this could explain this? I mean we are not talking about a matter of poor perception here. Having 35 seconds of 60 lost means the game would have presented itself as totally frozen, and even by an order of magnitude less than that it would have been very obvious with short skips every other second. The only conclusion I can draw from this is that the reported value doesn't actually mean much of anything and that it just isn't that simple.

In any case, it is actually awesome that you get a server this powerful for free and it is actually free it is not a scam. I believe they even guarantee to continue the free service years down the road... with servers it seems to be at a point now where it is like free bank accounts. The old banks still continue to charge ridiculous sums, while the new banks operate so efficiently they only profit from the free accounts which are making them more popular.

I still have my doubts that the Oracle Cloud thing will be viable. But ultimately we have to playtest this to know, and all this theory has just been nonsense. It totally survived the first 5 minute playtest.

I will post longer stats tomorrow.

As you can see, the OVH VPS experiences virtually no stolen ticks (peak value 20/minute), while the $0 Oracle VPS has much more users per machine and thus much more stolen ticks (peak value 8000/minute = 80 seconds per minute or 130%). OVH only had higher values when players were playing, but most of the spikes on the Oracle were occuring totally at random. I went to sleep at 7:00 so unfortunately we don't know yet if there will be more such spikes to values over 3000 and we don't really know yet if the game will choke or lag at those randomly occuring peaks (if they will even reoccur).

Even at 3500 stolen ticks reported per minute (which in theory corresponds to 35 out of 60 seconds of CPU time stolen, this theory makes no sense), the game ran just so smooth yesterday on the Oracle Cloud thing, that it was basically impossible to discern whether or not anything was wrong with it in the 5 minutes I played.

For the OVH VPS on the other hand this is pretty much case closed (as it has more than 100x less stolen ticks than that). It just works flawlessly, with no indication of anything differently.


I have tried to brush up on what is going on with the tick values, but it wasn't very fruitful. People say that it is normal that stolen ticks are misreported and poorly calculated, it depends on the situation.

I will continue to post statistics. But so far it just confirms that the reported stolen tick value doesn't really say anything about how smooth the game runs or how much lag you will have.

Just playtested this $0 Oracle Cloud thing with Cupatru, and damn, we both couldn't really tell the difference by a hair's breadth.

First he said the rocket guiding felt differently, but otherwise the server was fine. I had 2 quite obvious skips, but he and the others had not, so it was on my end. The rocket guiding seemed perfect to me, everything hitting where it should, etc. like yesterday. We then joined 3 other servers to compare it and played 3 minutes each. Cupatru said at first the guiding was better, maybe, maybe not, he didn't really know. I also thought at first that it seemed less sluggish visually somehow, but then it seemed even more sluggish than before. In the end by the 4th server it felt like I was seeing the 165Hz refresh rate of my monitor stutter and I believe he felt similarly that we had been overobserving things. Turned out for me that Thunderbird was using 100% on another core.

We also have been playing the $1 VPS from OVH, which already has been confirmed to work flawlessly.

Another playtest that was a great success for both servers.

We really should somehow get 30 players on the Oracle Cloud thing to see if it will hold at 90% CPU. Also there is still a very high chance that the Oracle Cloud will choke at certain times of the day or week. With the OVH VPS, I have personal and long experience that this has not been the case and the low stolen tick rate also confirms that the machines run pretty damn idle.

Server statistics now available!

http://xonotic.us.to/ - $1 OVH - ⚡⚡ Extreme Voltage ⚡⚡
http://vps-93234b88.vps.ovh.net/ $1 OVH - ⚡⚡ Extreme Voltage Duel⚡⚡
http://xonotic.oracle.us.to/ - $0 Oracle Cloud (x86) - ⚡⚡ ORACLE CLOUD - FOREVER FREE ⚡⚡ (edit: sucks, now offline)
http://193.122.48.73/ - $0 Oracle Cloud (Ampere/ARM) - ⚡⚡ NEW! ORACLE CLOUD - FOREVER FREE PLUS ⚡⚡ (edit: totally different hardware = awesome!)

It automatically displays stolen ticks vs load average over the last 24 hours on those pages. I will keep an eye on it but not be posting statistics here. Load values are higher, the more people play on the server. So this is how you can tell if the stolen ticks are totally random or merely related to programs running from your own system.

I just played on the Oracle Cloud server (x86) with 5000 stolen ticks per minute and it was no longer a smooth experience. 

What I experienced was almost identical to someone downloading stuff from my DSL line or watching Youtube while doing Windows Update, or something. Multiple short skips per minute, every now and then also a large skip with PL going to 20% and rarely also 60-80%. Then 2-3 minutes nothing, then again something minor then nothing again. I mean most of us are probably quite used to this and it might just be an everyday occurence, so for me emotionally speaking this wasn't even such a huge deal and I guess most people would have just kept playing regardless. This kind of stuff also happens on other servers sometimes, but rarely so. Lots of cheap VPS function MUCH worse than this, much much worse, so that it would be totally intolerable and makes you rage. This is not at all the kind of experience I had at 5000 stolen ticks.

But obviously in terms of server quality, this is not really something acceptable, and from an admin perspective it totally is a huge deal and no go. The only reason why I wouldn't call this server total shit now is that it really costs $0, and it is beefy and it really runs smooth "most" of the time at least it seems. 

Imagine this for example for Oracle Cloud x86:

• You want to run an exotic Xonotic mod that isn't really popular with many people, so it doesn't pay to rent a server for it. Now you can host it 24/7 with Oracle Cloud x98 for free and people can actually enjoy it and it will work very well most of the time, maybe even with high player count. If people play it rarely, they are very unlikely to join at the maybe exact 10-30 minutes a day where the server is choking
  
• You want to start hosting a server, try out new maps you made or certain configurations, but you are not really serious about hosting. Then the $0 variant is an excellent option as well.
  
• You can also use the $0 server if you want to quit hosting Xonotic, but you don't want to totally deprive the community of your server
  
• this server might also be perfect to host websites, or the map repository, at zero cost. It is an unmetered 600Mbit connection
  

Oracle Cloud x86 is not good enough for normal Xonotic server hosting, but it still could have many other uses.

Check out Oracle Cloud Ampere/ARM instead, which is just like light night and day 24GB RAM & perfect experience also $0.

The most tolerable is when you have less than 2k per day.

You seem to have confused days with minutes.

From the playtesting on the $0 Oracle Cloud (x86), it is true that a rate below 2000 stolen cpu ticks per minute indicates at least most of the time that the game runs smoothly, and it will not affect player experience. Whereas a value of 4000 often meant that the game was on the verge of developing PL and lag.

But I would say now after extensive playtesting, that stolen ticks is not a very usable/reliable metric at all.

One day the game was perfectly playable at 4000 stolen ticks on Oracle Cloud (x86), one day it were only 1000-2000 that made it almost unplayable.

I can't tell for certain of course why this is so, that day it might have been rather the network connection choking, so 4000 ticks could be an upper limit to consider. I believe that it is heavily related to disk IO, and having way too many customers on the same machine.

The $1 OVH VPS on the other hand so far only runs around 20-40 stolen ticks max when played, and it never had any lag and always ran flawlessly.

Ok here are the statistics! Please note that VPS will be the most overloaded on Saturdays, especially in the evening hours. So it is a pretty safe bet that it won't get any worse than that.







Again I can only stress the flawless performance of the $1 OVH VPS here, although it didn't see as much use today.

The Oracle Cloud's x86 performance was rather questionable today, and if this was any other day of the week it just would look rather bad as you can see.

BUT, BUT, this is only the x86 option of Oracle Cloud and they also offer Ampere/ARM processors. I will be testing those next.

Damn, we were just playtesting the $0 Oracle Cloud Ampere/ARM option and you get 4 exclusive cores for free with 24GB of RAM on a 600Mbit connection! It runs in a different datacenter too! In vmstat it constantly reports zero stolen cpu ticks, just like on a root server.

This would mean that VPS lag will be totally impossible at $0 with tons of beef in the server! You could even run 4 servers on this just like a root server!

From the CPU load it seems to go up to 40 players or more per core. 6 players caused only 13% CPU max. Not sure what the true ceiling is.

Damn this is such a great thing.

I wouldn't totally trust it without more playtesting, but at this point there is obviously very little that could still go wrong.

Edit: We played it for hours with about 6 people on a Saturday and it runs perfectly.

Edit2: It really runs like a $100 root server, absolutely perfect for Xonotic. I am pretty sure that no more playtesting is required.



http://193.122.48.73/ - $0 Oracle Cloud - ⚡⚡ NEW! ORACLE CLOUD - FOREVER FREE PLUS ⚡⚡

If you want to make one, here is my full server backup with cfg, maps, etc and instructions for installation in less than 15 minutes. or try my images for Xonotic on Dockerhub. Both work with ARM.

If you register on Oracle cloud, make sure to select a huge country as your "region" because you can't undo this selection, and small countries don't have ARM datacenters!! Maybe try to Google first if possible which regions have ARM and which don't. I selected Germany (NRW) but there are many others elsewhere too.

This map of regions is the best I could find. It doesn't show which regions have ARM, but maybe try to pick the ones with Azure, since those are probably the bigger and more modern datacenters.

Frankfurt Germany is often out of capacity, but not always. US Ashburn has never been out of capacity so far.

A clunk we discovered is that sometimes it keeps the old "boot volumes", thus blocking you from creating new boot volumes for free. Simply delete the old boot volumes under "block storage" to fix this problem. This happens if you create e.g. 4 test servers, then delete the instances, then create 4 new ones. It might explain how people had their servers deleted after the free trial because they were actually overbooking and the new servers were all using storage that would have cost money.

Malice told me to type status 1 in console. I'm not sure how reliable those stats are where it would matter (like what you would want to know most is long-term stats, whether or not chokes occur in the evening and weekends, which I don't think those stats can do). But looks to me they check out A+++.

host: ⚡⚡ Extreme Voltage ⚡⚡ - Ampere/ARM Oracle Cloud
version: Xonotic build 17:27:57 Jan 24 2023 - release (gamename Xonotic)
protocol: 3504 (DP7)
map: aneurysm_itemfix
timing: 6.5% CPU, 0.00% lost, offset avg 0.2ms, max 0.3ms, sdev 0.0ms
players: 3 active (32 max)

host: ⚡⚡ Extreme Voltage ⚡⚡ - http://extreme.voltage.nz
version: Xonotic build 03:46:30 Jan 24 2023 - release (gamename Xonotic)
protocol: 3504 (DP7)
map: cubes-x03-r02
timing: 5.4% CPU, 0.00% lost, offset avg 0.2ms, max 1.3ms, sdev 0.1ms
players: 2 active (16 max)

host: ⚡⚡ Extreme Voltage ⚡⚡ - DUEL & TDM
version: Xonotic build 00:13:08 Jan 15 2023 - release (gamename Xonotic)
protocol: 3504 (DP7)
map: quark_v4
timing: 4.3% CPU, 0.00% lost, offset avg 0.3ms, max 2.4ms, sdev 0.3ms
players: 1 active (16 max)

Updates on the servers: One of the $1 OVH VPS had over 100 stolen ticks per minute on the weekend for a short while. But playing at the same time was perfectly fine. The Ampere/ARM server from Oracle seems to be totally solid, better than the OVH VPS. Really impressive. No issues detected. There was construction work on the street though, and my DSL line sucks much more now, which makes further testing very impractical (I have to watch server and home line mtr while playing the game to be certain where lag spikes are coming from).

My verdict is that the ARM Oracle Servers are totally awesome and no issues detected.

Malice told me to type status 1 in console. I'm not sure how reliable those stats are where it would matter (like what you would want to know most is long-term stats, whether or not chokes occur in the evening and weekends, which I don't think those stats can do). But looks to me they check out A+++.

host:    ⚡⚡ Extreme Voltage ⚡⚡ - Ampere/ARM Oracle Cloud
version:  Xonotic build 17:27:57 Jan 24 2023 - release (gamename Xonotic)
protocol: 3504 (DP7)
map:      aneurysm_itemfix
timing:  6.5% CPU, 0.00% lost, offset avg 0.2ms, max 0.3ms, sdev 0.0ms
players:  3 active (32 max)

host:    ⚡⚡ Extreme Voltage ⚡⚡ - http://extreme.voltage.nz
version:  Xonotic build 03:46:30 Jan 24 2023 - release (gamename Xonotic)
protocol: 3504 (DP7)
map:      cubes-x03-r02
timing:  5.4% CPU, 0.00% lost, offset avg 0.2ms, max 1.3ms, sdev 0.1ms
players:  2 active (16 max)

host:    ⚡⚡ Extreme Voltage ⚡⚡ - DUEL & TDM
version:  Xonotic build 00:13:08 Jan 15 2023 - release (gamename Xonotic)
protocol: 3504 (DP7)
map:      quark_v4
timing:  4.3% CPU, 0.00% lost, offset avg 0.3ms, max 2.4ms, sdev 0.3ms
players:  1 active (16 max)

Updates on the servers: One of the $1 OVH VPS had over 100 stolen ticks per minute on the weekend for a short while. But playing at the same time was perfectly fine. The Ampere/ARM server from Oracle seems to be totally solid, better than the OVH VPS. Really impressive. No issues detected. There was construction work on the street though, and my DSL line sucks much more now, which makes further testing very impractical (I have to watch server and home line mtr while playing the game to be certain where lag spikes are coming from).

My verdict is that the ARM Oracle Servers are totally awesome and no issues detected.

Have you tuned your servers? I mean edited sysctl, installed irqbalance and tuned daemons and/or something like? I still think most of the lags is a network issue but tuned server can be smoother.

I have not tuned them. If there are issues with VPS it is totally external, you can't help it.

What lag do you mean?

The only server that had lag was the Oracle x86 variant (but not ARM) and this is definitively *not* a network issue.

Lags like teleportation, also I've experienced invisible projectiles or skipped sounds. May be some of that can arise due to server overload but I'm agree with you - reasons of most of bad things placed outside datacenter. Server tuning is not only about lags, it's about overall game sense - a bit less and stable ping, a bit less latency. I played on some servers where movement was like in a liquid or just weird but ping looked ok. Guess tuned server can work better, even if it gives you few percents. But idk how to measure influence. Mb just log pings while playing and compare with/without tuning?

Ok you are talking about lag on other servers, cheap VPS and bad datacenter would be the cause of that.


Concerning Linux kernel, sysctl and such there really are no tuning options available, zero. Nothing you can do makes any amount of difference, however small it might be.

This is because virtually all of Linux tuning focuses on scheduling and disk IO, and it only works to a measurable degree if there is a lot of it. Xonotic runs entirely from RAM, so there is zero disk IO to interfere with the game. And scheduling only is important if you are running multiple cpu-intensive processes at the same time. The default scheduling is already pretty damn well optimized for usual tasks, such as running 2 or 3 Xonotic instances and a webserver simultaneously. So on a normal server, it is not like messing with the scheduler would improve anything, rather the contrary.

There are also a lot of tuning options you can do as a hypervisor (=VPS provider), this also only targets disk IO and scheduling, but as a VM/guest there are zero. 


In short: For an exclusive high quality server, the only place where tuning could take place is in Xonotic config.


But then I am not sure what kind of tuning you are referring to, and there are only very few options that could make a difference (like sys tickrate). The standard options should already be optimized though. You could make a case to double the tick rate for duel servers, which could result in a slightly more fluid experience. But I don't think any server actually does this, and maybe it would just feel slightly off and irritating if you changed this value, rather than provide any subjective improvement (especially when not raising multiples but fractions).

You are also not the first one who is talking about "latency" or "delay" in addition to ping. Players have told me this on servers that I was playing on and I could 100% confirm from my own client that there was no such issue with the server. You have to realize that in Xonotic, ping is measured directly from CPU time not only from network time (or at least so I was told). This means that the only additional latency can come from your device, such as monitor or mouse and the entire latency/delay from the server to your own game is simply called ping. But I have observed what you are talking about myself quite often actually. One time we were playtesting and I noticed that the game felt sluggish, despite normal frame rate and ping. I found out later that my Email client was hung up and using 100% CPU on the other core, which caused this issue. From your description, it felt the same way or very similar to say the least. Since this issue doesn't occur on all clients, I think it is very fair to say that this "sluggishness" or extra "latency" with normal ping is by all odds not a server issue and probably just caused by hung programs/webbrowser running in the background.

Concerning the tuning: Maybe if you could specify exactly what in your view would qualify as "tuning", we could run some metrics on that?


Also if really such "secret" tuning options existed that aren't already default, they should be part of the documentation.

Thanks for detailed answer. I think I need to refresh my mind after heavy tasks at work =) Really Xonotic is not so heavy (while we haven't 20+ players on single server) and all magic happens between server and client. Anyway I have a python script for measuring ping while playing written by andreymal. He checked our pings in-game for few servers last year and showed me results. For example


 It would be interesting to dig into that. I'll do some test with tuned daemon profiles like network-latency when I have free time. My opinion: there may be a small ping improvement. Ofc there may not be =)

Ok you are talking about lag on other servers, cheap VPS and bad datacenter would be the cause of that.

So, thank you for your research and testing.

How does Oracle run now? Is it alright for your purpose?

How many players have you tested with? 20+?

Do you run the webserver httpd on the same server as xonotic?

How much bandwidth do you get? What about lag?

ty

/FAF

Just played on host: ⚡⚡ Extreme Voltage ⚡⚡ - Ampere/ARM Oracle Cloud as a single player with bots... worked smooth as butter.

Wonder how 10v10 with nobots would be???? Have you stress tested it?

Thanks for detailed answer. I think I need to refresh my mind after heavy tasks at work =) Really Xonotic is not so heavy (while we haven't 20+ players on single server) and all magic happens between server and client. Anyway I have a python script for measuring ping while playing written by andreymal. He checked our pings in-game for few servers last year and showed me results. For example


 It would be interesting to dig into that. I'll do some test with tuned daemon profiles like network-latency when I have free time. My opinion: there may be a small ping improvement. Ofc there may not be =)

I tried to tune Feris multiple times, with all kinds of profiles, but I ended up removing it all and reversing to defaults without tuning at all. I don't know what it was, but I believe that people were "loosing their shots" after some time, on one of the low-latency profiles, though I didn't investigate it further just removed the altercations to see if it got better, which it did. I use VirtIO drivers on the nic btw and they perform the best. I really don't have the resources right now, to fine-test the advantages of tuned-profiles on the kernel, I'm happy now as it is per default with 20 players hammering.

I think the best thing would be to make a Xonotic distribution on Just-Enough-OS or something like that. Make it a very light no-bloat distribution with recent packages. Who would be the maintainer? - or is Debian 11 minimal satisfactory - nevertheless, could be cool with a finished distribution with all kinds of enhancements, like a browser Admin-CP, install-scripts, map-installer, filebrowser, etc.

Ok you are talking about lag on other servers, cheap VPS and bad datacenter would be the cause of that.

So, thank you for your research and testing.

How does Oracle run now? Is it alright for your purpose?

How many players have you tested with? 20+?

Do you run the webserver httpd on the same server as xonotic?

How much bandwidth do you get? What about lag?

ty

/FAF

The ARM variant runs *perfect*, there is no other word to describe it. Just like a $100 root server, very beefy, no chokes, no stolen CPU time (it has dedicated cores) nothing wrong with it at all.

I think the most players we had were 11 or so, and it didn't even go above 15% CPU I think. Unfortunately we just don't have enough players around for playtesting. If you could somehow round people up on a Friday evening or Saturday, that would be great. But since it runs on dedicated cores and it has a 600Mbit connection in an ARM-exclusive datacenter that few people even use ... I think such stress tests are kind of unnecessary at this point. We already know from the sheer CPU power it has, that it should support over 40 players or so. And it doesn't share that CPU power with anyone. So the only choke can happen at the network, but like I said little people even use this data center, so there is no issue there either.

Webserver runs on same server.

No lag.

Btw. check out what I wrote about tuning to Xord. Tuning just doesn't work because Linux is already totally tuned to run as a virtualized guest. Otherwise all tuning options are limited to disk IO only, but Xonotic runs entirely from RAM during a match and literally has zero disk IO. This is why there cannot be IO-related chokes in Xonotic on the ARM server, even if there actually were any. If you are planning to use the ARM server for stuff like video hosting and cloud storage, be aware that I haven't done any serious testing against disk IO issues for this reason. It seems to be fine, sometimes sluggish in the SSH console like on an expesive VPS and maybe slightly worse on Friday and Saturday evening like usual. It was funny, on the x86 variant Xonotic would even run smooth when the console totally locked up for 2 minutes at a time, because it could not write to bash history. Like I said it runs entirely from RAM. But of course no such thing ever happened on the ARM variant.

Ok you are talking about lag on other servers, cheap VPS and bad datacenter would be the cause of that.

So, thank you for your research and testing.

How does Oracle run now? Is it alright for your purpose?

How many players have you tested with? 20+?

Do you run the webserver httpd on the same server as xonotic?

How much bandwidth do you get? What about lag?

ty

/FAF

The ARM variant runs *perfect*, there is no other word to describe it. Just like a $100 root server, very beefy, no chokes, no stolen CPU time (it has dedicated cores) nothing wrong with it at all.

I think the most players we had were 11 or so, and it didn't even go above 15% CPU I think. Unfortunately we just don't have enough players around for playtesting. If you could somehow round people up on a Friday evening or Saturday, that would be great. But since it runs on dedicated cores and it has a 600Mbit connection in an ARM-exclusive datacenter that few people even use ... I think such stress tests are kind of unnecessary at this point. We already know from the sheer CPU power it has, that it should support over 40 players or so. And it doesn't share that CPU power with anyone. So the only choke can happen at the network, but like I said little people even use this data center, so there is no issue there either.

Webserver runs on same server.

No lag.

Thank you. Yes, lets hope the best for the free tier solution, which I hope will become more and more common. We need free servers

Let's find 20 people to test the server.

BTW: Which region are you using on Oracle? Did you get access to Germany on the free tier?
BTW2: Yes, the kernel is already tuned to VM. That was my conclusion too.