At first, thanks a lot for your responses. According to infos I found in the list archive I changed the values for net.inet.ip.ifq.* as follows net.inet.ip.ifq.maxlen=1024 (256 * number nics) net.inet.ip.ifq.drops=0 net.inet.ip.ifq.len=0 This had no effect on the network throughput.
In the next step I increased the value for net.inet.tcp.recvspace and net.inet.tcp.sendspace to 262144 and that had big impact on the network throughput. The network throughput increased from 550Mbit/s to 940Mbit/s. Because of this effect I would like to know if there is detailed documentation availbale, describing the sysctl values and there effects? Unfortunately the manpages don't describe these things detailed. Another issue is based on the disk subsystem. Not the disks in the openbsd router, but the disks used by the clients. At the moment maximum writing speed is about 54Mbyte/s which is half the network speed. I guess the only thing I can do here is to use raid0 to increase read/write speed. nstag, 12. Februar 2008 15:40:31 schrieb Stuart Henderson: > On 2008/02/13 01:04, Darren Spiteri wrote: > > Try tweaking this sysctl: net.inet.tcp.recvspc > > > > Give it sysctl -w net.inet.tcp.recvspace=262144 and run your tests. > > Tweak it down from there. > > This is irrelevant on a firewall/router. > > > > I have been using openbsd as router and firewall for several years without > > problems to do routing between four networks (ethernet, ipv4) using 100 Mbit/s > > nics. After upgrading network infrastructure to 1000Mbit/s the network > > throughput is not as expected. I expected a throughput around 100 Mbyte/s and > > Check net.inet.ip.ifq; search the archives for more about this.