eu server

Discussion in 'Planetary Annihilation General Discussion' started by elodea, February 27, 2015.

  1. elodea

    elodea Post Master General

    Messages:
    1,694
    Likes Received:
    3,040
    seems borked from overseas. many lags

    anyone can confirm?

    * 02/03/15
    Just wanted to update on the OP in case anyone was confused by this thread

    Server issues, from AU to EU and US anyway, are all good. It was a one off thing that Jables suggested it might have been.

    As for EU and US to AU server, that might still be a problem maybe. I'm not a position to confirm that.
    Last edited: March 2, 2015
  2. violetania

    violetania Well-Known Member

    Messages:
    395
    Likes Received:
    330
    Everyone in BRN that's Australian has had a lot of lag from the EU... lately i've been getting a lot of lag from the US which i didn't use to get. haha.
    elodea likes this.
  3. cola_colin

    cola_colin Moderator Alumni

    Messages:
    12,074
    Likes Received:
    16,221
    playing on AUS servers from germany is horribly laggy as well since ever. I doubt that is something Uber can solve
    elodea likes this.
  4. jonasmod

    jonasmod Active Member

    Messages:
    234
    Likes Received:
    190
    had big problems with us servers yesterday. eu servers were fine though. i live in sweden.
    lapsedpacifist likes this.
  5. mayhemster

    mayhemster Well-Known Member

    Messages:
    394
    Likes Received:
    425
    Is there a way from in-game to work out the IP address of the server?
  6. cola_colin

    cola_colin Moderator Alumni

    Messages:
    12,074
    Likes Received:
    16,221
    Using the UI debugger yes, if you add a debug log in the right place.
  7. lapsedpacifist

    lapsedpacifist Post Master General

    Messages:
    1,068
    Likes Received:
    877
    Same here, lots of lag on US servers yesterday, and normally US servers run fine from the UK. I actually had some server lag with EU though later on (but not as much) so could just be a dodgy connection at my end.
  8. jables

    jables Uber Employee

    Messages:
    812
    Likes Received:
    5,537
    I believe PlayFab was working on things yesterday which may have caused a bit of this. If it persists please let us know.
    elodea likes this.
  9. fuzzels

    fuzzels Member

    Messages:
    90
    Likes Received:
    34
    Just use the custom PA Stats servers, then you can pick a 'local' one.

    There are two servers in EU:
    http://planet.unfix.org/paservers/

    @cola_colin needs to update his US box (at least, I gather that the Phantom one is his ;)

    To use these, install PAMM and then from there install PA Stats and they will pop up in your server browser.

    Would be cool if PA had an integrated way of 'donating' servers to the community in a similar way so that everybody playing PA could benefit from this. (hint hint @jables :)

    As my nephews moved to GTA5 recently (8,10,14, oh yeah... ;) they stopped playing Minecraft, hence I might convert that box in a secondary public PA server.
  10. exterminans

    exterminans Post Master General

    Messages:
    1,881
    Likes Received:
    986
    I doubt that you can provide the same class of hardware as it is currently used for the Uber provided servers. Right now, there is no hardware available on the market which could provide any more significant speedups, all you can do is to wait for an optimization pass on the software side.

    Better save that box for private 40 player games, but don't expect to much, performance wise.

    When you really want to provide additional capacities, better cancel that server and buy commander skins ;)
  11. fuzzels

    fuzzels Member

    Messages:
    90
    Likes Received:
    34
    > I doubt that you can provide the same class of hardware as it is currently used for the Uber provided servers

    As it is unknown what kind of hardware they use, how are you able to make such a statement?

    > Better save that box for private 40 player games, but don't expect to much, performance wise.

    Works fine for that, but also running multiple 4v4 etc games. Performance is excellent actually.

    > When you really want to provide additional capacities, better cancel that server and buy commander skins

    There is no 'canceling' those servers, they are properly racked up in a datacenter; it is not just a random VPS I am using for them....

    I think you are assuming people are using low-performance VPSs as private toys... not the case...


    + (the edit): the more important thing of this is that servers are distributed much better around the world and thus have lower latency times than the playfab provided servers. There are few 'cloud' companies that have servers really close to everybody.
  12. exterminans

    exterminans Post Master General

    Messages:
    1,881
    Likes Received:
    986
    Amazon EC2 hardware, probably the old C3 instances (2,7Ghz Ivy Bridge eight core + HT, dual socket system with 60GB ECC RAM). One instance hosts multiple PA servers to achieve better utilization as peak and average load differ greatly with PA. It's unclear whether they have rented full instances (c3.8xlarge / c4.8xlarge) or just the smaller VPS.

    Recent or precise numbers haven't been published, but about 2 years ago they said 4 server processes per instance (that number might have changed due to optimizations), and the servers are still hosted in the Amazon cloud, so only the numbers may have changed.

    And yes, the Amazon data centers are quite well distributed and have great uplinks, less than 30ms latency to the nearest data center from about all large ISPs, not included the latency on the last mile from DSL or alike.
  13. fuzzels

    fuzzels Member

    Messages:
    90
    Likes Received:
    34
    Yes, this might sound like a e-penis thing but the primary thing is: there are people who would love to contribute their servers to the pool to make PA better to play in various areas where playfab is not too performant yet.

    > (2,7Ghz Ivy Bridge eight core + HT, dual socket system with 60GB ECC RAM)
    [..]
    > One instance hosts multiple PA servers ...

    At 4 processes per instance, ignoring overhead that thus gives a dual-core 2.7Ghz CPU with 64/4 = 16GiB mem each.
    Then my silly E3 haswell has that matched quite easily.

    > less than 30ms latency

    I have 10ms to my boxes; note that is not "30ms latency + the latency for the last mile", that is the _total_ latency from my desk to the machine in the datacenter; both IPv4 and IPv6 same latency for that matter as both are native.

    The boxes are then also only 40km away from here, you know in case something breaks and one needs to get to the datacenter quickly to fix the hardware. (software can be fixed by using the OOB management ;)


    Heck, please realize that in Europe we have latencies typically of <30ms on standard home links even to _other countries_ (then again, countries in Europe are typically smaller than US States). Eg I reach hosts of mine in Amsterdam from home in 22ms, from the colod boxes it is about 15ms (different transit providers and thus paths).

    Note that Amsterdam to NYC is about 80ms. Thus 30ms is almost half of that...
  14. mayhemster

    mayhemster Well-Known Member

    Messages:
    394
    Likes Received:
    425
    The OP was about bad lag on ranked matches where we don't have the ability to choose the server we play on. The lag wasn't the normal delay you get from the distance to the server but was actually far worse with game clients locking up for long periods of time. I saw this a number of times this weekend on twitch streams and also people playing me in matches also reported it (although I didn't get any myself).
  15. fuzzels

    fuzzels Member

    Messages:
    90
    Likes Received:
    34
    Hence, the option of setting up a local server to share with other players. This is possible with PA Stats, though would be cool if it was integrated, also because then they could be integrated into the ranking system (though, if the server op is evil they could make some people lag or other benign things... meh).

    "Lag" can be a multitude of things. Being down-under one has many network disadvantages unfortunately: pricing/throttling and other nasty things I heard about next to the big spiders ;)

    Could be the client giving issues or the network connection having problems or the server. traceroutes would be a good start, and so would at least be client CPU graphs and memory load. Then again if you are playing from AU on a EU server.... that is 300ms at least.


    I can only assume that PA has some kind of internal diagnostic to determine client or server cpu/mem/io load and to diagnose network problems...
  16. exterminans

    exterminans Post Master General

    Messages:
    1,881
    Likes Received:
    986
    East Europe / Asia would be such regions, but I don't suppose any of us could afford collocation with virtually unmetered traffic (and stable upload rates of >50Mbit / server) in any of these locations.
    That calculation is flawed in multiple ways. First, the over-allocation causes the system to behave entirely different from 4 individual systems. The average match would surely do fine with a dual core, but it can't handle the border cases like 10 player matches and alike. That's when you actually do need all of the hardware. Sure, it can still go haywire when two or more server instances happen to missbehave on the same machine, but the risk of that happening is significantly lower than a smaller machine being overburdened with a single large game.

    Oh, and I think you missread the numbers. Eightcore + HT + dual socket makes for a total of 32 threads respectively 16 physical cores. It's most likely also more than "just" 4 server processes running per instance nowadays, since back then when the statement was made, the biggest machine available did feature only 8 physical cores.
    Even in countries like Germany, a huge number of home links is far above 30ms. ADSL2 behaves rather shitty when used in environments with lots of interferences, so you can get easily an additional latency of 20-40ms from that alone, I have seen up to 60ms as a baseline. With Fastpath enabled, via cable or FTH, sure, that's a different thing. With FTH, I can reach Frankfurt in 7ms as well.

    And even if it wasn't for the trouble on the last mile, routing in Europe is a mess as well. There are several large ISP which are rather restrictive about peering and will happily deroute your traffic via the US. Especially during top hours, when their willfully undersized local peering ports are overloaded. German Telekom for example....
  17. fuzzels

    fuzzels Member

    Messages:
    90
    Likes Received:
    34
    Which exact part of "East Europe" and "Asia" are you meaning, these are rather large areas with quite different network connections to the rest of the world. Better question: at which ISPs are the clients having issues?

    Hence my "ignoring overhead" point.

    Thus your argument is that those 4 instances on one box is not adequate either.

    I think you misunderstand how virtual machines work and what kind of limits/restrictions one can put on them.

    Ignore HyperThreading, everybody building real servers do that. HT is only useful if you mix a lot of operations, which is common for graphically intensive stuff, but not for just pure server things.

    Thus you are stating that they are even more overloaded ;)

    That just has to do with the two modes of DSL in germany: low latency or highbandwidth. Typically DSL is configured as the latter and thus one gets high latency (and high latency is something you never want, not even for browsing, as the Internet is bufferbloated as it is...)

    Pick a better ISP by voting with your money. It is not like these countries do not have a multitude of cheap options.
  18. exterminans

    exterminans Post Master General

    Messages:
    1,881
    Likes Received:
    986
    These limits are not applicable to the Amazon EC2 instances. Amazon performs no over occupation at all and your VM is pinned to physical cores. You are also not charged for CPU time, but for the uptime. In return, new instances may be allocated at any time, only as long as you actually need them.

    If you want to make efficient use of your instance by maximizing the average load, you have to do so yourself.

    Playfab has probably a set of dedicated instances (which come at a lower rate) for the base load, and rents additional instances on the fly whenever deemed necessary.
    No, it's not about overhead at all.

    It's about the random distribution at which the load occurs. I'm going to use a simplified example, ok? Let's say that you have a 10% chance that one process requires twice as much resources. If you were to scale the hardware accordingly so that it could handle these 10% as well, you would waste about 40% of your resources on average. Now if you scale the hardware x5, and you run 4 independent processes in parallel, you have a chance of ~95% of successfully running all 4 processes without throttling while having only about 5% of the resources underutilized on average.
    Scale that up even further and the utilization increases even further, while you can even handle scenarios where the resource consumption does not only double, but increases by even larger factors.


    Besides, it's not as if latency would matter (much) in PA either way. Even with 200ms, the game still feels quite responsive for most tasks, only exception being stuff like manually dodging projectiles where you rely on feedback from the simulation. After all, the server is doing only 10fps either way, so you already have a 100ms jitter in there.
  19. fuzzels

    fuzzels Member

    Messages:
    90
    Likes Received:
    34
    I think you are mixing "VM instances" with "PA server instances".....

    > These limits are not applicable to the Amazon EC2 instances.

    Of course they are applicable. If you put two VM instances on a box and they both have dedicated set of memory and cores assigned they won't affect each other. Disk IO and network IO is a separate problem though.

    > Amazon performs no over occupation at all and your VM is pinned to physical cores.

    Exactly what I am stating. Thus if you have a 2x8 core box, and 4 VMs on it, they do not affect each other too much, though there is a little bit of overhead and disk IO and network IO are still a shared resource that cannot be 100% perfectly separated.

    > Let's say that you have a 10% chance that one process requires twice as much resources.

    I noted above, you are mixing up "VM instances" and "PA server instances". If you decide to run your PA servers inside the same VM you are losing the separation and then indeed, one process can steal away resources from another. Though, even there you can pin processes to a certain amount of cores in the same way that KVM/qemu does it when distributing cores to VMs....

    The CPU/mem/swap/disk/net schedulers will actually have more information if you do a 1:4 VM-PA instances compared to 1:1 VM-PA instance though.

    > Even with 200ms, the game still feels quite responsive for most tasks,

    Europe to Australia is at _minimum_ 350ms round trip. Hence having a more local server can definitely be beneficial.

    Note also that if you have 200ms network latency and the sim is another 100ms later, you are at 300ms already, as they do add up.
    Though, for a FPS that will hurt a lot, not too much with a sim like PA where a move command or a shoot command does not directly mean hit/miss especially with rockets flying around...

    edit: VM:pA (thus V M : P A) becomes a smiley, changed that to VM-PA in the above text
  20. elodea

    elodea Post Master General

    Messages:
    1,694
    Likes Received:
    3,040
    Just wanted to update on the OP in case anyone was confused by this thread

    Server issues, from AU to EU and US anyway, are all good. It was a one off thing that Jables suggested it might have been.

    As for EU and US to AU server, that might still be a problem maybe. I'm not a position to confirm that. If you're playing from AU, you should change your server to US anyway :p.
    jables likes this.

Share This Page