Koby Botick
Eighty Joule Brewery
Goonswarm Federation
2
   |
Posted - 2011.09.10 12:51:00 -
[1] - Quote
Malcanis wrote:[quote=CCP Veritas]Why are you wasting resources fixing lag when you should be fixing lag ffs?
"Fixing lag" indiscriminantly is impossible. Fixing lag for 1000 players in one system may be possible.
The reason this is a serious and depending on some factors impossible to solve problem lies in the information theoritcal domain. The problem, extremely simplified is this: If you have 100 ships in one space and one ship of those houndred performs an action, you have to communicate this action to 100 clients (99 others and to the one initiating the action so the client knows it's done ok). Imagine that's sending 100 messages or 100 letters to everyone for that single action. Now if not 1 ship does something but everyone does something, that's already 100 times 100 messages to send. This communication problem is an n^2 (n - squared) problem.
For people with knowledge in performance optimization in software everything is clear at this point, for all others: n^2 means that the more people you have there, the faster the amount of communication grows, and it grows quadratically. The main problem is you can only add more hardware or better hardware linearly. If instead of one big server, you put in two or four big servers to handle the problem, you only help a bit, since the problem increase outgrows the processing power you add. Because the problem grows quadratically, while your solution can only grow linearly (at best). Worse, the more you add the less increase you get.
This is a core information theoretical problem that is impossible to solve. The trick is to be handle practical limits fine and mitigate the odd outliers. Which is precisely what CCP is doing at the moment. Note that improvements until lag appears can be done, because that depends on a lot of factors that are controllable, including the efficency of the software handling the issue, however there is simply a limit on how much you can do and at some numbers you can't do anything at all, since you reached the limit where the quadratic growth of the problem just takes off and goes through the roof. |