Microsoft Research TechFest - Using P2P to speed up multiplayer gaming (and other things)

prencher wrote:


The download works perfectly, its the streaming one thats barfed.

Excellent interview, though what about NAT related P2P issues? Most people don't have inbound ports opened, so how do you get around that?

Haza!  That's where Vista can comes in with it's Teredo technology, which is an ipv4 to ipv6 transition technology.  It allows for automatic NAT traversal if you are hooked up to an IPv4 gateway.  If you have Vista you can try this out in the Window Meeting Space application.  It's a p2p app that uses teredo and lets you transfer files and also share desktops directly without needing to configure the NAT.

I thought the focus groups where really smart.  People complain a lot that they're getting killed because their connection sucks.  If the bots are utilized to much people might say the game is somewhat unfair.  The focus group is a great way to fix the problem. 

I just though of something that could apply to distributed gaming too.  In my networks course we were talking about multicasting that is in development for applications like IPTV.  With something like IPTV with traditional unicasting the TV server would need to send the video to each and every client.  If there are millions of clients, the server upload bandwidth quickly becomes the bottleneck.  One solution is IP Multicasting where there is packet replication and branching at the network level.  Essentially a tree is constructed where the server is the root, internal nodes are routers and the leafs are clients.  The root server would send video to a few routers, these routers would further send to other routers and eventually this branching tree would get to the client.  More packet duplication occurs starting the periphery of the network so the overall bandwidth is more evenly distributed.  I can see this applying to multiplayer games really well since it seems the upload capacity is the bottleneck.  If a player is sending state update info to everyone there is a large unessecary duplication of packets at the source when this duplication could be done closer to the clients at different routers.  Combining this with the type of technology mentioned in the video  it would seem you could have even larger sets of coexisting characters while also relying less on bots.  Too bad multicasting seems so far away in terms of depoloyment.  Well at least people can look forward to the tech mentioned in the video.

Chadk wrote:


Btw. we already have MASSIVE multiplayer games. Eve online have up to 33k players online at peak times. Its really amazing.

But if objects are managed by the conneted peers, wouldnt there be a potetional problem where the unit would change the actual location of the object that it manage, giving the player an unfair advantage?

These massive multiplayer games aren't really massive because the players do not coexist at the same time in the same place.  It works because players don't reside in the same local place and also because the game root servers are run on large server farms that have much higher bandwidth than end systems.  With shooter games one of the clients acts as the server which means the server's processing power and bandwidth are the bottleneck.  Shooters can also use p2p like in the video, but then again the bottleneck is the upload of each peer and the amount of coexisting players.

The problem with inconsistency came to my mind too.  It seems the focus groups does a good job fixing the problem by allocating more bandwidth to characters that are in focus.