> As for 1.7 beta - had a strange issue. I only received 1 node, after about 10 minutes, > in DHT when I usually get about 200+. I right clicked and disabled DHT, right clicked > again to enable DHT, and it did the (Login) and numbers starting going up. Another > application start I had no DHT issue. Should be fixed since build 1167, if this is the duplicate ID issue. The behavior is similar > XP SP2, Server 2003 and Vista (besides Home Basic) are capable of 10 half open > connections. 2000, XP and XP SP1 are capable of something like 65,535 half open > connections (anyway, a really high number, at least what I wrote). I won't blame you at all for writing that, because Microsoft did publish that "fact" regarding this feature in SP2 and other OSs. However, that fact is so badly out of real context and is misleading, that it should be considered false. Connections attempts are serial. They would be limited by the datarate of the network card and the speed of your upload connection -- not to mention that you only have about 3800 ephemeral ports that you can use, and limited memory in both Windows and your NAT device. > The thing is that it > should be higher than that, at least 32 or 64. Why should it be higher than that? When you exceed the outgoing half-open rate of X per second, Windows writes a warning in the log and imposes a limit of X per second. Then it merely delays the connection attempt by the amount necessary to stay within the limit -- no Winsock error is returned to the application due to this delay. This delay should only happen when starting a torrent or getting an announce with a large amount of new peers. With everything except Home Basic, even the largest "simultaneous" attempt might incur a delay of only a few seconds. I am concerned about the new limit of 2 per second on Vista Basic and possible impacts on P2P apps generally. Winsock still shouldn't issue an error when 2 per second is exceeded, but if applications have tiny timeouts, outgoing connection attempts in a newly (re-)started large torrent or after a large announce with peers may be timed out by the application before the SYN packet is actually sent by the network stack. It shouldn't cause a fault, failure, or severe error. Anyway, 2 seems rather draconian.