Should I Migrate to VPS or Dedicated to Resolve TPS Issues

Discussion in 'Hosting Advice' started by penguin4215, May 17, 2019.

  1. Hi there,

    I am currently hosting a server with Bisect Hosting, on a 10GB budget plan. This has worked decently well for my server for the few weeks I've been with them after I switched from Skynode. However, recently I've been having major issues with TPS (timings listed below). After almost a week of troubleshooting and discussion with the host and the server hosting community we've been unable to resolve this problem.

    I am looking to switch to a dedicated server, particularly OVH's game server range as it is the only one I could find within my server's budget. I've heard mixed things about hosting with OVH from reviews dating back a few years so I wanted people's thoughts on switching to a dedicated server from a shared hosting provider, and in particular to OVH. If anyone has any other dedicated server recommendations I would be glad to hear them. If you feel that switching to a dedicated host would in fact not resolve my TPS issues, please let me know too.

    Timings: (I am aware I have a large amount of entites, but the TPS drops don't track very well. A couple days ago, I had 3.2K entities at a flat 20 TPS, while the next day I had 1.6K entities at 13 TPS.)

    Any help is appreciated.
  2. Tux


    Consider further lowering your entity-activation-range (Paper has an entity-activation-range for water mobs, which you should probably lower), which should reduce the amount of time spent doing entity ticking.

    As for moving to OVH... well, you're on a shared host that uses OVH, technically you're already using them. For a small server, a dedicated server is probably overkill - a VPS might be a bit better for your needs, but it depends on what you're trying to do with your server.
    • Like Like x 1
  3. BisectHosting and SkyNode are both companies which resell others.
    OVH is a budget provider and you can not expect high-quality services from them.

    As mentioned before with SkyNode you are essentially already using OVH, I'd recommend you to look into a better shared hosting solution or VPS. A dedicated server is not needed unless you want to host more than 200 players.
    • Like Like x 1
  4. Unless you're hosting a network with hundreds of players a dedicated server most likely isn't needed and OVHs offerings are most likely overkill and overpriced for your use case. I agree with the sentiment that better shared hosting or a VPS would be better for your needs.

    However if you really really want a dedicated box then I'd wholeheartedly recommend renting a VPS from Hetzner's server auction. Just make sure to set the min CPU Benchmark to 9000 and tick the SSD/NVMe option. Any of the servers listed with mentioned requirements should fit your use case and will definitely be an upgrade from your current shared hosting without breaking the bank.

    If you're considering a VPS then I wouldn't recommend Hetzner or OVH because neither actually specify the CPUs they're using, also they're selling vCores which might be anything from 1-100% of an actual CPU core of the CPUs they're hosting the VPSes on. The Deploys Classic Virtual Servers seem like a decent offer and actually specify the CPUs used, though I don't have personal experience with their VPSes, only the shared MC hosting.
    • Like Like x 1
  5. A dedicated server will perform better than shared hosting, but that's not necessarily the right solution to your problem, especially since you have only 40 slots. That shouldn't be enough players to cause problems with shared hosting, and you certainly threw enough memory at the server.

    Your web page says you have "huge mob farms and grinders", and that might be the real problem. Looking at the timings, I see 78 million entityTicks for rabbits, which indicates that someone has 2000 rabbits. If they're all in one place, that can drive your TPS down into the toilet. Note it's not the actual number of entities, but what they're doing.

    So you can go with a more powerful dedicated CPU (not the low end ones), or you can do something to limit the size of mob farms/grinders, the latter of which goes against the stated goal of your server of "no gameplay-alterting plugins". Along those lines, what do you have specified for the gamerule maxEntityCramming? Note that gamerule is set per world, so check them all.
    #5 Bobcat00, May 17, 2019
    Last edited: May 17, 2019
    • Like Like x 1
  6. Thanks everyone for your input. You’re definitely right that’s I shouldn’t go for a dedicated server I’m aware my servers scale does not need it, I’ve just been desperate to find a solution. If you have any recommendations for VPS providers I could look at in future that would be appreciated.

    As for Bobcat00’s reply,

    Entity cramming is currently set at 8, which is already lower than I’d rather. As for my server, the premise is that players are free to do what they want so I don’t want to limit them by acting against large mob farms. That being said, I will run my world through the region fixer tool when I get the chance to remove any concentrated areas with entities.

    However, for the hardware and memory I have on my server. I would have thought that the shared hosting situation should be able to handle ~2k entities total, I didn’t think that was unreasonable at all?

    I would be happier to upgrade my hosting to a higher performance VPS with a high clockspeed CPU and would rather do so than limit my players. Again, any recommendations are greatly appreciated. :)
  7. Devouch for server auctions. They're old, decommissioned dedicated servers that people ordered many years ago. The best CPU you'll find is an ivy bridge. I did a bit of quick maths using this chart, and you'll have about a third better performance on an 8700K than the best CPU you'll find on a server auction. In addition to this, the drives will be on the brink of death. My friend bought a server from the server auction, drives had 49406 power on hours, which means its performance is very much degraded, and could fail at pretty much any time. In addition to this, Hetzner's network isn't the best. Hetzner's DDoS protection isn't great, and I get 30ms ping to my Hetzner box, whereas I get 20ms to a VPS I have about ~300km away - their routing isn't the best.
    • Like Like x 1
  8. - You'll also pay more than a third higher price for an 8700K CPU. I understand if you were coming forward with better offerings of equal price but at least I don't know of an 8700K offering that's anywhere close to the ~30€/m for the 4770 + SSD storage offerings Hetzner has. If there is such I'd love to know so I could recomment that to people!

    - Yes the drives are well used like most server drives are but any peformance degradation shouldn't be an issue with SSDs and there's always two so RAID1 covers the unlikely situation that one of the SSDs fail. And yes even with SSDs there's a heightened chance of both of the drives failing in a short period of time if they're used up but it's such a miniscule chance (with the SSDs) that it's unlikely to happen. Also that's the situation where backups are your friend.

    - Their routing isn't the greatest and definitely not as good as Deploys but it'll do well enough, especially for a survival server. As for the DDoS protection; a year back @FusionStyleFX from Deploys claimed Hetzner DDoS protection isn't good and that his friend has access to booter that could down my server in a bit. I told him go ahead and try, he tried (got email notifcation of the attack) and nothing at all happened to the availability of my server. It might not be the best DDoS protection but based on that I'd trust it myself.
    • Like Like x 1
  9. The 8700K was just to put the ivy bridge CPU I found on the server auction into perspective. Before I switched from OVH, I was on a SYS GAME-2, which had a 4790K (~15% better performance than a 4770), 32GB of RAM and 240GB SSD (which pretty much brand new, not on the brink of death like Hetzner server auction drives) for £35/mo. Also note that people could have been mining on the CPU, which will degrade performance (CPU performance degrades over time, either way). So basically, on OVH, you're getting a better CPU and newer hardware that has less use.

    Right but then you have another issue - the drives could be near their write cycle limit, giving you the same problem. There's also the problem of EEPROM taking a relatively long time to actually erase data from the disk, as whole blocks of data must be copied to another block, before the original black can be erased. Therefore, a fresh SSD where the blocks simply aren't touched yet will always perform better.

    Which will lead to downtime, especially if it fails on a Friday night or weekend. Hetzner don't generally respond to requests on weekends. Only time I've had a request handled out of hours is for my box on my friend's account, who puts about $300/mo through Hetzner, and is a "valued customer" or something, so he gets some extra benefits.

    If someone is writing a lot to the SSDs daily, which is common in server settings, i.e. seed boxes, they are very much likely to be on the brink of death. It can be automatically assumed for pretty much 99% of server auction boxes that the drives are pre-fail.

    Doesn't mitigate the downtime, and you'll have rollbacks, depending on how often you take backups. If you're on a server auction box, then you probably won't be buying a storage server, and you'll be taking backups manually, probably once per day at max, probably once a week.

    Last I heard, Hetzner null route servers when the attack gets above a relatively small amount, and you have to email to get net enabled again, and obviously, you can't do that on Friday night / Saturday / Sunday
    • Like Like x 1
  10. I'd go investigate what's going on with all the rabbits. 2000 entities is not a problem. But what are they doing with all those rabbits? If I'm looking at the timings properly, your server is spending half of the tick time dealing with rabbits.
    • Agree Agree x 1
  11. Yes you're getting slightly better hardware that's less used but you'll be paying about a third more for the SYS offering. It's obviously @penguin4215 job to figuire how much he's willing to invest and what's valuable to him.

    Where have you gotten the idea that CPU perfromance degrades over time in any sort of meaningful way from?

    Obviously the drives being heavily used would be an issue with applications where you continue to do a lot of writes to the drive, degrading it more. But since the amount of writing to a drive when hosting a minecraft server is so low I'd say there's a diminishingly low chance of having a drive fail and a practically nonexistent one of having one after the other fail within the time perdiod Hetzner would need to replace one worn out drive.

    And if one is so paranoid about the SSDs failing then you could order one of the auction servers, check if (both of) the drives are heavily written to and if they are cancel the server as per Hetzner's 14 day cancellation policy, doing this until you find a server with at least the other drive being a fresh replaced one or just not written to all that much.

    This would obviously be completely valid point if we were talking about hard drives but we're not so I don't think the drives are an issue.

    I (unfortunately, kinda) don't have experience with Hetzner's parts replacement service for their auctioned dedis so I can't comment on the speed of the service. Though even if the replacement time would be a week I'd still stand by the statement that the chances of the other drive failing in that timeframe is practically nil.

    Nothing would stop one from taking incremental backups to an off-server location every 10minutes even, if you were paranoid enough.

    A year back it was good enough to handle a booter @FusionStyleFX thought would suffice to take down my server's connection, which for me is sufficient proof that even if it's not the best it's definitely enough for any kind of small survival server project someone might be running, like I did and like @penguin4215 does.
    • Like Like x 1
  12. A third? Where did you get that from? £35 = ~€39, 39 - 33 = 6. 6 / 33 * 100 = 18%. 18% more for newer, better hardware that doesn't have degraded performance.

    Running at high temperatures damages the hardware. This is why when buying a used GPU, you look for one which hasn't been used in a mining operation. The same logic applies to CPUs. Evidence:
    First 2 are about mining, which is the same as running any application which puts the CPU under a lot of load.

    Auction servers are either retired servers, or custom servers that people ordered in the past and cancelled (this is really expensive, you'll only really have companies doing this). The company using the server before you was definitely not running a Minecraft server.

    ... which still leads to downtime

    lol. Do I even have to respond to this one?

    Except it is a valid issue, because both drives will stop being writable at the same time, meaning you'll have a bit of data loss, and you'll also have downtime. There's also the issue of erasing with EEPROM taking a long time.

    Or, for another 6 euros, you can buy a server which won't give you a week of downtime?

    Already covered this in my previous post:
    If you're on a budget server from a budget provider already, you probably won't have enough money to be buying a storage box too.

    It's well known that Hetzner takes a long time to actually initialise mitigation (in which time, your server will have dropped all active connections). Look online, you'll find tons of horror stories. If someone actually wants to down your server, they'll have no problem.
    • Like Like x 1
  13. [​IMG]

    45€ / 28€ = 1,6 meaning a 60% higher price for the GAME-2 compared to a comparable offering from
    The GAME-1 being it would be 34 / 28 = 1,21 so even that's 20% more expensive

    Albeit the GAME-1 would probably be quite sufficient for OPs needs.

    AskGN which is the most credible of your sources (in my own opinion) just talks about the degradation of CPU cores casuing them to require hihger and higher voltage to maintain a certain clock speed or not being able to run at X at Y voltage anymore because of degradation of the CPU from heat or voltage. This would in the case of an overclocked CPU cause either the performance to decrease because you can't run the CPU at a higher clockspeed anymore or you need to increase the voltage at said clockspeed which will in itself degrade the cpu more and generate more heat which will also degrade the cpu, eventually leading the CPU to degrade to a state where you can't sustain said clockspeed anymore

    However the 4770 isn't overlcocked and even if it was used for Monero mining or some other CC you can mine with a CPU I doubt it would've been degraded enough for the CPU to experience issues in maintaining its original boost clocks.

    My point was that even if the drives were at the brink of failure relative to their lifetime write limit (before complete failure), for example 585TB/600TB (97,5%), you'd have to host a really really active Minecraft server for a really really long time to actually have the server write rest of the 15TB until failure of the drive.

    No, you don't have to respond to anything.

    I repeat myself: A year back it was good enough to handle a booter @FusionStyleFX thought would suffice to take down my server's connection, which for me is sufficient proof that even if it's not the best it's definitely enough for any kind of small survival server project someone might be running, like I did and like @penguin4215 does.
  14. I used to pay £35/mo for my GAME-2. SYS always has sales, with discounts that last for life, so you purchase during a sale.

    The clock speeds of CPUs fluctuate. For example, at idle, all 4 cores on my 6600K sit at around 800MHz. This is so the voltage can be decreased, as higher clock speeds require high voltage. Why does this happen? To prolong the life of the CPU (as well as reduce power consumption + temperatures). This is why boost and base clocks exist. A CPU cannot sustain the boost clock forever, as it requires extra voltage to do so, which damages the CPU (as well as it heats up, which damages the CPU too), so therefore, a CPU will drop back down to base clock after a few minutes (or if it begins to thermal throttle).

    6 years solid of being under load isn't enough to degrade performance?

    I've had my EX42-NVME for a couple months now, just over. I ran smartctl when I first got it, and the drives were brand new (apart from install OS + Hetzner's drive tests). I ran smartctl again:
    Code (Text):
    Data Units Written:                 4,815,223 [2.46 TB]
    I'm writing about 1.25TBs per month, just from hosting one VM, an MC network, Redis, MariaDB, a Discord bot, a Rust server and a Gitlab instance.

    Good, because I can't even fathom the logic behind what you said.

    I was witness to an attack on a Hetzner box about a few weeks ago. A $20 booter took it down with an L4 instantly, Hetzner mitigation took over 5 minutes to kick in (for a volumetric attack! really?). Although the server did technically have net back, it was taking ~30 seconds to respond to simple HTTP requests for a static page (no, this wasn't an L7 attack on the web server, L4 on the whole box, couldn't access SSH for example), meaning Hetzner was letting the majority of requests through still.

    If you want DDoS protection for a Hetzner box, get a BuyVM KVM slice in Luxembourg, and GRE tunnel through it. It's about 10ms between BuyVM and Hetzner (may vary with which DC you get, Nuremberg is a bit closer than Falkenstein), but the reduction in ping you'll get through the Voxility routing to BuyVM should make up for it.
  15. It's hard to say if Hetzner blocked the attack or the booter simply was not really strong.
    Hetzner had a few infrastructure upgrades and their DDoS protection is said to be a bit better though, but I don't really think it comes near what OVH offers, or even Voxility.
  16. Well if the CPU would have been ran at 100% load for the last 6 years then sure it maybe can't run at its boost clock for too long before it has to take the clocks back down. But I doubt the CPU having been used at 100% for the last six years to mine Monero or some other CPU mineable coin.

    Also you should probably choose one: The CPU has been ran at 100% for 6 years causing performance degradation or the drives have been completely used up throughout the years as a seedbox or whatever because I doubt there's more than a few servers on the auction which have had both the CPU used for mining at 100% all the time and both of the drives used up to their limit.

    To me it seems quite simple and straightforward

    Could we just talk in € or £ but not both.

    I don't have any sort of perspective on the drive usage of the other applications listed so I can't really comment on this one.

    Well having to get a VPS for tunneling and a storage box or something equivalent for backups takes to price of the whole ordeal up enough that then it wouldn't make sense anymore to not just go to one of SYS' offerings since you get good anti-DDoS and some backup space included in the price.
  17. Mining isn't the only thing that's CPU intensive. Data processing, transcoding, compression, graphics processing and mathematical computations are all examples of CPU intensive tasks.

    Alright, how about a Plex server? Constantly transcoding video for multiple users is very CPU intensive. Data is also constantly being read from the disk, and written if you're using Ombi (which you should be).

    Alright then.
    • Hetzner will refuse to sell you any more servers once they notice you're just buying and instantly refunding them
    • Refunds can take days to be processed. I'd rather just buy a SYS GAME-2 and spend a little more rather than wait 2 weeks trying to find a server with half decent drives

    Was using pounds, because that's what I paid for my SYS box with

    It was just a little side note, hence separated from the rest of the paragraph. I prefer Hetzner because you get more resources than OVH for cheaper, but I need this $5 server for DDoS protection, just sharing what I do.
  18. Thanks everyone, I appreciate the discussion.

    At this point, It looks like for my use case and budgeting, I would be best off waiting for a SYS sale for the Game-1 dedicated server. Alternatively, I’m open to suggestions for quality VPS hosting, as long as I would have enough CPU performance and RAM to prevent TPS issues in future.

    As for my current shared hosting server’s performance, the TPS drops are consistent whenever more than 30 players are online. I regularly check the timings report and it looks like it is never one entity type responsible for the issue, and the highest average tick usage changed frequently. Considering the fact that the hardware of the shared server that I currently run on should be enough to handle 30 players and ~2k entities just fine, I am worried that upgrading to a VPS or even a low end dedicated server would not actually solve the problem.

    I’ll post an updated timings report when the server next peaks so that the issue is as obvious as possible.
    #18 penguin4215, May 19, 2019
    Last edited: May 20, 2019
  19. this is your main issue, if one person falls asleep at their PC (even if you have afk kick) they can just setup an afk pool machine that lets them bypass your afk checks. a few hours of that and your server will crash, it only takes one person, especially with 2 gb mem.

    If you want to let them have large mob farms without restrictions get a mob stacker/merger plugin. You can also switch to paperspigot and set mob spawner mobs to not have AI.
  20. @goflish, I’m not sure if you’ve read the majority of this thread, my replies or the OP. The shared server I currently have has 10GB of RAM, which should be more than enough to handle ~2.5k entities. In addition, you’d also see on the timings I am already using Paper.

    From the timings, I’m not convinced it is 100% a problem with entity count, considering the fact that the TPS drops do not match up with the number of entities present on the server. The only thing that correlated with the TPS drops is the playercount.