By using this site, you agree to our Privacy Policy and our Terms of Use. Close
SvennoJ on 01 June 2013
Edit Thread
youarebadatgames said:
SvennoJ said:

But how would cloud compute solve the problem of 800 man fleet battles? It's the client struggling to render all the data that's coming from the server. Or your connection not being able to keep up with the amount of data the server is sending.
MMORPGs mainly have a limited player size to protect for overcrowding. Throwing everyone together on a distributed server system is not rocket sience, not having 10k people come to the same area so all the clients crash is the problem. Like the good old days of staring at the floor trying to navigate the Bazaar in Everquest, look up and the sight of hundred's of player traders brought any client to it's knees.

The only solution is to switch to an OnLive/Gaika type solution where the data stream and effort on the client side are constant at the cost of a bit of extra lag.

In EVE it's not the client that is the bottleneck, it's the server because of the computations of lots of ships, physics, modules, etc. going on at the same time.  Maybe true in other MMORPGs, but EVE is pretty light client side, and in fact EVE will slow down time server side to keep up with the calculations.  Just having a distributed server model might have made it go more smoothly to scale up demand when there is a huge spike in system.

Ah I did not know that.  http://penny-arcade.com/report/article/planning-for-war-how-the-eve-online-servers-deal-with-a-3000-person-battle Interesting they can temporarily move solar systems to another server and slow down time to 10% to deal with massive battles. Yes great candidate for cloud computing. I wonder what hardware they use, 3000 actors doesn't sound that many to keep track of. I guess with everyone shooting missiles it starts to get pretty taxing.

It's a complex issue to split up into parts though. I guess you could first divide all objects into a binary quadtree to manageable numbers, process each quad in the cloud, then redo all the objects that crossed quad boundaries. Perhaps it already works that way, at some point all the extra back and forth communication will outweigh the gain from enlisting external computing power. Which leaves slowing down time to solve the bandwidth problem. Sounds like a fun challenge to work on.