EmbryonicRsts between two Linux boxes
Subsystems (2)
Hi,
I have three Linux boxes on the same physical network. One runs a web server (webfs) and the other two run an HTTP traffic generator. Under high load, with many simultaneous open connections (~1000) one of the clients gets a TCP RST ("read: Connection reset by peer").
netstat -s on the server box gives:
...
TcpExt:
3002 resets received for embryonic SYN_RECV sockets
...
Which, if I understand correctly, means that the server TCP stack receives a SYN|ACK or a RST|ACK from one of the clients for a SYN_RECV socket.
How can this happen?
Thx.
PS: there's 100ms network latency between the server and the clients (emulated by NistNet). And I want to saturate the server. That's the reason I need so many simultaneous connections.
Signature · 1 line
Eric