Re: [glob2-devel] Finalized Design Idea for YOG

[Leo Wandersleb]

Stéphane Magnenat wrote:
> If you want to go for tcp, at least try to do it intelligently: first try to 
> find a player in the game with no router/nat and sufficient bandwidth.

hard to do i guess.
also i guess as most people think the host has a special role i guess we should 
give them exactly that: the initiator of the game is the router.
if not possible, yog does the routing. 
also there should be a clear message to the players about who is the router.

> If 
> this player exists, use it as a temporary router. If no such player exists, 
> you can fallback on yog for routing.
>
> I continue to think it is non scalable solution but I won't refrain you from 
> doing it, but I won't host this yog.
>
> It could also be good design to separate yog meta-server from yog routers. In 
> fact, it could be a good idea to spawn one router per game. I even already 
> have code for such router, especially if you use the new generation messages 
> I sent to the ml some days ago.

as i have massive stability concerns the more we put into yog i think having 
one process per game is not a bad idea. also because these can then be on any 
server. what i think is important is to have one central yog service where all 
players meet in the first place. they should then be able to choose where the 
router should be spawnd by checking the according player or one of the 
available servers. we then could even publish a daemon anybody could install on 
his server that reports to the central yog service to be able to spawn new 
routers.
this daemon should know if routing capacities are free and quantify that in 
some way so only the top 5 routing options show up for each game.
also a test-phase (those countdown 5 s) should be used to determin if the 
spawnd router really works for all clients. if not, the countdown stops with a 
message disabling the according server and reporting the issue to the yog 
service.

> Do not remove voip nor make it optional: make it work on Windows instead.

testing it results in what i got several times when i accidently hit "v":
glob2: VoiceRecorder.cpp:139: int record(void*): Assertion `readLength == 
toReadLength' failed.
i put it on the bug tracker and stuck there for an hour now. sorry for the 
delay ;)

greetings, Leo Wandersleb