Multiplayer - back and forth, old and new
#11
(05-31-2014, 06:24 PM)aqqman link Wrote:Loni and RPN are from Brasil as far as I remember which is located in south America which is far away from Europe, and closer to US, so I find it a bit awkward to call them Euro players.
I guess if it is so wonderful we should expect mass return of real Euro players to the Rangers. Will see.

Sorry, in my mind everyone with a high ping is in Europe, apologies for the awkwardness. Post edited.
Reply
#12
(05-31-2014, 08:44 PM)aqqman link Wrote:... involves individual routing paths between clients which in some cases could be even worse than using host relay.

While this is possible, in practice the Internet should always route your UDP packet more efficiently than routing it through an arbitrarily chosen middleman; especially when you consider the addition of software/CPU induced lag on the descent host.


Reply
#13
The host-induced Host-lag is mostly what I've been tinkering on in the past months. And yes, I can easily see how the latest 0.58.1 build could triple the ping compared to P2P.
Like aqqman had suggested so many times it looked as if the Host would be just overloaded with packets it could not handle. Especially since the CPU usage seems to increase with new players (which I am currently optimizing on). While technically it could handle this, the symptom of how a program handles the packets was comparable as the host would process the packets in a client-fashion time-frame instead of doing it "asap". And since aqqman also suggested to make a separate threat for this, I did it in a very similar fashion. Not a separate threat but a separated event loop that calls network functions in much smaller time-intervals.
I am now on a current average of 2,5ms which the packets might reside on the host - compared to 30ms to 120ms in 0.58.1.

In 0.58.1 you can "simulate" this delay by simply slowing down the whole program using -maxfps where lower values would drastically increase the ping. With the current code base you can try the same thing but the ping for each client will not change.

That really solved a big bottleneck of which I am - if I may be so bold - very proud of.

Granted I could only test the results over connections within Germany which does not compare to the U.S. at all I admit. Which is of course also why I would like you guys to test it and gimme your honest opinions.
The greatest pleasure in life is to do what people say you cannot do.
Uhm... Honey, there's a head in the toilet!
Reply
#14
(06-01-2014, 01:26 AM)spud link Wrote:(...) in practice the Internet should always (...)
You mean 'in theory'.
(06-01-2014, 01:41 PM)zico link Wrote:I am now on a current average of 2,5ms which the packets might reside on the host - compared to 30ms to 120ms in 0.58.1.

In 0.58.1 you can "simulate" this delay by simply slowing down the whole program using -maxfps where lower values would drastically increase the ping. With the current code base you can try the same thing but the ping for each client will not change.

That really solved a big bottleneck of which I am - if I may be so bold - very proud of.
That sounds really cool - I want it - bring it in  ;D
Reply


Forum Jump:


Users browsing this thread: 1 Guest(s)