[Twisted-Python] High traffic connection issue

Trying to do some scalability testing on my first twisted server. When I have hundreds of connections and moments of very high server activity I am seeing that clients are unable to connect. Here are some numbers: 250 open TCP connections. I see 857 operations in my log during the second where a new client tries to connect. In the client log I see that connectionMade() is called. I am pretty sure I don't see a corresponding connectionMade() on the server side (could this be right?). Then, about a second later I get a connectionLost() on the client with 'Connection to the other side was lost in a non-clean fashion'. Is this to be expected? Should the client just be expected to retry? Thanks

On 05:47 pm, ted.pederson@gmail.com wrote:
This depends a lot on what your server does, as well as what platform you're running it on. If all your server does is accept TCP connections and ignore data received from them, and you're on Linux and using epollreactor, then I would expect you to have no problems up into the tens of thousands of connections range. On the other hand, if you check an RSA signature for every new connection, and you're on Windows using win32eventreactor, I would expect failures once you get to a few dozen connections. Jean-Paul

On 05:47 pm, ted.pederson@gmail.com wrote:
This depends a lot on what your server does, as well as what platform you're running it on. If all your server does is accept TCP connections and ignore data received from them, and you're on Linux and using epollreactor, then I would expect you to have no problems up into the tens of thousands of connections range. On the other hand, if you check an RSA signature for every new connection, and you're on Windows using win32eventreactor, I would expect failures once you get to a few dozen connections. Jean-Paul
participants (2)
-
exarkun@twistedmatrix.com
-
Ted Pederson