On Sunday 16 December 2001 13:48, Oskar wrote: > On Sun, Dec 16, 2001 at 01:26:28PM -0500, Tavin Cole wrote: > > On Sun, Dec 16, 2001 at 02:05:27PM +0100, Are wrote: > > > My 34 mb logfile contains: > > > > > > 8675 entries of > > > Time (Freenet.node.Node, Thread-#): Error getting initial state for > > > message: source: Freenet.ConnectionHandler@# msg: QueryRestarted @ # > > > Freenet.node.BadStateException: This message is a response > > > at Freenet.Message.getInitialState(Message.java:96) > > > > This suggests that a lot of chains are getting pushed out of the message > > store, so that when the QueryRestarted is finally received the node has > > already forgotten about that chain and tries to process it as a new one.
This is troubling. > > > > We could increase <messageStoreSize>.. but it is already 500. So that > > means the node is regularly in a situation where it has 500 pending > > request chains (apparently many of them hanging around because the > > request is timing out somewhere upstream). This sounds bad... We need better rate limiting. My node sometimes gets up to 100 inbound requests per *minute*. There's a positive feedback loop here. Nodes get overloaded so they handle requests more slowly so the network gets more congested so nodes get more overloaded. The load average on my public node machine sometimes exceeds 10. > > This is what keeps surprising me about the current 0.4 network. There > are just too many requests - I have even suspected that it is > continually the target of a continual DoS attack from somewhere, but > something else seems more likely (how about the initial discovery > requests? Could we temporarily tag those in the request ID to see if > those make up a large part of the traffic?) It might also be the never > terminating restarts which could keep requests going for days. > > Maybe we ought to build a private debugging network where we have > greater insight into what is going on. > > <> The htl decrement on Restarted bug was only fixed a couple of days ago. There are probably a lot of nodes on the network running the the old broken code. Also, it looks like there are 2 to 3 times as many transient as non-transient nodes. Here's what I base this on. I have seen ~= 140 unique nodes make inbound connections to my machine since the inbound contact logging code was put in. My node never has more than about 35 outbound good noderefs. I wouldn't be surprised if there's someone DoSing the network. It might not even be malicious. I might just be someone running scripts to keep their content in the network a la the 0.3 "Artificial Inseminator". --gj -- Freesites (0.3) freenet:MSK at SSK@enI8YFo3gj8UVh-Au0HpKMftf6QQAgE/homepage// (0.4) freenet:SSK at npfV5XQijFkF6sXZvuO0o~kG4wEPAgM/homepage// _______________________________________________ Devl mailing list Devl at freenetproject.org http://lists.freenetproject.org/mailman/listinfo/devl
