Bug still exists...

Hi.

The vm slow-down bug starting at 4000 sockets still exists. I now have finally traced the issue down. It depends on the number of sockets AND the bandwidth you need. So I suppose it depends on the number of packets flowing through the NIC. I managed to completely remove the bug by using two cards and the nexthop-parameter to use eth0 and eth1 with the same gateway-ip. Maybe interrupt-problems? Slow-down appears on intel and broadcom cards. Other cards not tested.

Chris