[ENet-discuss] Reliable Sequence Numbers overflowing

Lee Salzman lsalzman1 at cox.net
Thu Oct 11 22:19:38 PDT 2007


Okay, after some more testing there are some cases where, if you have 
some 8K packets incoming reliable packets at the right time (during a 
sequence number wrap-around), it increases the LIKELIHOOD of the clients 
disconnecting, because it causes the receiver to drop the packets more 
often than usual until the sequence numbers fully roll over. Normally 
this would cause no problems because the other send would just keep 
resending until they went through, but if you're bombarding the other 
end with new packets the resends might not get through and, again, 
increase the LIKELIHOOD of disconnecting.

It's a pathological use case, and its not really a bug in terms of how 
ENet handled it, but I made the handling of it in CVS somewhat more 
robust now so that it can stand the insane bombardment of packets 
better. :)

So if you're brave, you can try the current CVS and see if that does any 
better a job at handling it... if it wasn't a simple starvation error as 
described in the previous email.

Lee


More information about the ENet-discuss mailing list