Re:Erlang and First-Person Shooters in online games [pdf]
by cageface on Sunday June 19, 2011 @11:44PM
(#2672756)
I'm surprised Erlang isn't used more often. Out of the box it seems to solve a lot of the hard scaling problems many business face now. Maybe it's just too exotic.
I wonder if they could have been as successful with Scala + Akka. They mention a desire for at least limited static typing at the end of the slides. In my experience well-defined, statically typed interface declarations can eliminate a lot of dumb bugs.
Re:Erlang and First-Person Shooters in online games [pdf]
by glassx on Sunday June 19, 2011 @10:10PM
(#2672582)
Nice write up.
Apparently the consensus is that Erlang is amazing for 'lower-level' infrastructure code but not so great for high-level business code. I wonder if Reia and Elxir might play a role in changing that...
Re:Erlang and First-Person Shooters in online games [pdf]
by getsat on Monday June 20, 2011 @03:28PM
(#2675610)
I'm sure they could have done everything in Erlang, but they mentioned that they only have four developers that do Erlang within the company. Erlang developers aren't as prevalent as Python/Ruby developers.
Re:Erlang and First-Person Shooters in online games [pdf]
by thesz on Monday June 20, 2011 @06:59AM
(#2673658)
I thought they implemented game network code, scalably. That would be cool. So I was a little bit disappointed when read that they only keep stats using Erlang.
Re:Erlang and First-Person Shooters in online games [pdf]
by thesz on Monday June 20, 2011 @09:40AM
(#2674091)
I should clarify a bit. I think that contemporary FPS need real-time game server logic. When I saw link title I thought they finally ported some game code to Erlang.
All stuff you mentioned has nothing to do with game logic.
It does not require hard real-time response like one required from network game server.
And yes, I already know that I could use Erlang for non-hard-real-time tasks.
Re:Erlang and First-Person Shooters in online games [pdf]
by sbalea on Thursday June 23, 2011 @08:50AM
(#2687634)
Akka is attempting to bring as many Erlang features as possible to Scala. They're doing a great job, but imho they won't be able to go all the way due to JVM lacking certain features
I wonder if they could have been as successful with Scala + Akka. They mention a desire for at least limited static typing at the end of the slides. In my experience well-defined, statically typed interface declarations can eliminate a lot of dumb bugs.
Apparently the consensus is that Erlang is amazing for 'lower-level' infrastructure code but not so great for high-level business code. I wonder if Reia and Elxir might play a role in changing that...
Other than that, that's pretty interesting.
Slide 19 lists far more than just stats-keeping:
> Our core server for controlling Python
> – Managing 100,000s of concurrent TCP connections
> – Scheduling/queuing of tasks for python
> – Metrics gathering (SNMP)
> – Presence server(fragmented mnesia)
> – Message passing (nb: this might be the "in-game messaging" service mentioned in the presence server slide 21)
> Other standalone game-related servers (expanded on slide 25)
> – Transient in-game data (seems to be a k/v store used for e.g. dynamic chat channels)
> – Testing bandwidth (estimation of client bandwidth via blasting UDP packets to the server)
> – Ranking leaderboards (real-time rankings, >15m users, built on ets and a customized gb_trees)
All of this is listed under "How we use Erlang"
All stuff you mentioned has nothing to do with game logic.
It does not require hard real-time response like one required from network game server.
And yes, I already know that I could use Erlang for non-hard-real-time tasks.