Offline Play step 1 (LAN play) update is live!

Discussion in 'Planetary Annihilation General Discussion' started by jables, October 9, 2014.

  1. tatsujb

    tatsujb Post Master General

    Messages:
    12,902
    Likes Received:
    5,385
    this.

    while I support going the "results" over "theory" route, any real "results" with private users will hosting servers will come from multithreading.

    as private users are more likely to have big cores rather than racks.
  2. doridian

    doridian New Member

    Messages:
    9
    Likes Received:
    12
    Me and my friend actually got a dedicated server running on a third machine.

    We ran it on a ubuntu machine with a small workaround so the server wouldn't just listen for localhost, you then can connect to the server using the ip join ui component mod.

    The first client to connect will take ownership of the game hosted.

    It's not exactly pretty but it worked pretty good.

    PS:
    Start params we used are: --headless --allow-lan
    Twinstar and cptconundrum like this.
  3. cola_colin

    cola_colin Moderator Alumni

    Messages:
    12,074
    Likes Received:
    16,221
    lies, the 16 core amazon server we just put a 32 player game on peaked at more than 800% cpu, so more than 8 cores at full load.
    I guess you actually have one sim thread and a pool of threads (as the server in fact logged "INFO Starting background thread pool with 14 threads") that does other things like manage per player data and whatnot?
    You're half way there. Now just split the sim thread, playing at a sim speed of ~2 fps was no fun xD
    I'd really like to read a more detailed post about where server performance is going. I am kinda afraid the only way to really scale well would be pretty good multithreading for the sim, as all big servers one can get just have a ton of "okay" cores.
    That's a pretty hard to do, I know, but there should be a bunch of things that can be done nevertheless :S
    Twinstar likes this.
  4. maxcomander

    maxcomander Active Member

    Messages:
    328
    Likes Received:
    129
    Ive been having a little play with the server, and know that Ive got rid of any bandwidth issues, It looks like the next bottleneck is sim speed (as others have said)

    I think for know overclocking the cpu will give best improvement, but if it was possible to paralelise the sim over more cores, that would be the ideal solution.

    That said, I'm sure it's not that easy. If it was, it would probably have been done already.

    Seeing as manufacturers seem to be focusing on core count over ipc, maybe this would be a worthwhile investment for the pa engine...

    Again this is something I would be happy to contribute to via kickstarter or other funding channels.

    Edit.. I was also wondering weather getting someone like intel onboard as maybe a sponsor or something, seeing as atm the isn't much point in getting anything above a highly clocked i5 if your computer is solely for gaming.
    Would have thought Pa would be the perfect game to advertise the power of high end i7's.
    Last edited: October 12, 2014
  5. Gossy

    Gossy New Member

    Messages:
    11
    Likes Received:
    8
    Having now had a crack at LAN play, it is so smooth its amazing, I live in aus and the lag meant that anything over a single planet battle was atrocious after about 25 minutes (and I have a good connection). This game now pleases me greatly ^_^
    spittoon and carn1x like this.
  6. carn1x

    carn1x Active Member

    Messages:
    389
    Likes Received:
    156
    I agree, I'm in Hong Kong on 500mbit and it still ran like a turd. Playing offline now is fantastic especially with the recent patrol optimizations :D
  7. MrTBSC

    MrTBSC Post Master General

    Messages:
    4,857
    Likes Received:
    1,823
    afaik they once had a deal with intel regarding MNC ... it didn´t seem to go well ... i don´t know the details though ..
    maxcomander likes this.
  8. squishypon3

    squishypon3 Post Master General

    Messages:
    7,971
    Likes Received:
    4,357
    Had to do with an amazing new GPU and ray tracing. With true real reflections, and not just another rendering of the game.
    maxcomander likes this.
  9. maxcomander

    maxcomander Active Member

    Messages:
    328
    Likes Received:
    129
    Yes I recently found out about this MNC was going to demonstrate the ray tracing capability's of larrabee but then Intel decided that the chip wasn't ready for the discrete gpu market or something...

    I seem to remember Nvidia demonstrating some ray tracing around that time, possibly had something to do with intels decision.

    Still I'm sure the cash from intel came in handy ;)

    Here's the article (just in case you havn't seen it yet).

    http://www.polygon.com/features/2013/3/19/4094472/uber-hail-mary-monday-night-combat
  10. MrTBSC

    MrTBSC Post Master General

    Messages:
    4,857
    Likes Received:
    1,823
    yea have readen it before .. was a while ago though .. but still thanks
    maxcomander likes this.
  11. maxcomander

    maxcomander Active Member

    Messages:
    328
    Likes Received:
    129
    I have been testing the server using the offline capability and can assure you that it doesn't use more than 7-8 threads. This is also including running the client.

    I did get one anomaly were I saw 12 threads in use, but that had to be a bug as I've been unable to replicate it despite running bigger games.

    The server seems to perform very well with highly clocked cpu, client sim speed seems to be the next bottleneck.

    Running a game with 8 large planets and 9 absurd ai's only started to to slow down after about an hour or so and it was the sim (at around 4.5fps).

    I do agree with you about multithreading the sim, however I'm gussing that to do so will take a ton of money and resources, even if it it's possible at all.

    Best chance we have imo to play those HUGE battles were all dreaming about is to get uber more money,

    Kickstart Hr.

    Buy all dlc for pa.

    Buy mnc.

    Convince all the hater's there wrong about uber, so they start buying Uber products.
    Last edited: October 18, 2014
  12. cola_colin

    cola_colin Moderator Alumni

    Messages:
    12,074
    Likes Received:
    16,221
    I've done multiple tests on many core servers by now (up to 24 cores, but sadly so far only up to 32 human players)

    I am quite sure that if you throw even MORE players at it it will use more than 8-9 cores, as it seems to do *something* per player. I've definitely seen the server alone use a peak of 900% on a 16 core dedicated server with 32 players.
    We should make a 50+ player game just to prove this suspection, but I doubt the server uses a fixed amount of 9 cores, as it does log something like "I just started a threadpool of n-2 threads" where n is the number of cores you have. So the 24 core server i.e. starts 22 threads. If we create enough work for this threadpool via a lot of players it should end up with 22 cores at full use + the sim thread + an unknown other thread

    With the sim I simply agree with you. It seems to be a single thread. At least one thread per planet should be feasable though, hopefully a little more. Scaling with more cores is simply put the only way the game really can scale with big servers. Big servers don't have super powerful single cores afterall. They have dozens of average ones.
  13. maxcomander

    maxcomander Active Member

    Messages:
    328
    Likes Received:
    129
    Unfortunatly I cant test that until I can host online. but I have seen 60% loading on my cpu and 12 threads being used simultainiosly , I just cant replicate it. You would have though using ai's would be more cpu intensive than real players, and that's what I based my thinking around.

    If you have any idea's or mods I might try I'd be happy to give it a go. I love tinkering with this game=)
  14. cola_colin

    cola_colin Moderator Alumni

    Messages:
    12,074
    Likes Received:
    16,221
    Well read my thread about dedicated servers in the wip modding forums.
    I have not made any tests with AIs at all. No idea about those.

Share This Page