Re: [GNUnet-developers] p2p 'massively multi-player games (MMGs)'

[Chris GauthierDickey]

> So, for your latency question, I would have to say that "it depends  
> on what do
> you want to do". See, there are really no implementation for it at the
> moment, or something for us to test that. What we have is in  
> GNUnet's roadmap
> one application called "gnunet-chat", that does exactly that: swaps  
> messages
> between nodes. When that's built we'll be able to see the real latency
> challange in GNUnet, but, otoh, we believe that gnunet-chat <will  
> work>, and,
> depending on what kind of stuff you're wanting to do, that can  
> already be
> enough. Most MMOG's send little more traffic than a simple chat system
> does... Unless you try to use the network to send graphics data.

Most multiplayer games are aimed to take advantage of the available  
hardware
that players have. The networking is tweaked to prevent the bandwidth  
from being
overloaded and the servers do a few fancy tricks to send only the  
most important
data if the data to be sent is more than what the player's bandwidth  
can handle.

In essence, most MMOs will target network utilization for around  
80-90% of a 56k
modem (we can see this from measurement experiments). I think we can  
expect
to see this increase as the population that has broadband increases  
to the
majority of players. So I wouldn't say MMOs send little more traffic  
than a simple
chat system (though it'd be impressive if someone could chat at  
48kbps!).

The main issue here is the topology of the network. As I stated  
before, we have
limits on latency given a particular type of game. We can probably  
assume that
bandwidth will continue to increase at the end-user over time, so  
bandwidth is
not as much of an issue. With games, it's about latency, update  
frequency, and
security. If the topology of the network is not conducive to fast  
message
dissemination, then it's fairly easy to show that it won't be able to  
support a particular
type of game.

If you look at the current research in this area, many of the P2P  
systems use
structured overlays, such as Pastry. This introduces a logarithmic  
number of
hops from point A to point B. However, these are end-system hops, not
network level hops, and therefore cause a significant amount of  
latency that
make the most common type of commercially successful games unplayable.

On the other hand, a pure P2P network with all-to-all message passing  
is clearly
not viable either. One of the things I'm looking at is building the  
P2P network
based on the locations of players in the virtual world. Thus, instead  
of randomly
organizing players into the network, where a message between two  
players that
are close in the virtual world might have to traverse the entire  
network, the message
will instead take only a single, direct hop to the other player, or  
at least minimize
the number of hops required.

Chris