On 2009-09-11, Tobias Sarnowski <[email protected]> wrote: > I am running an OpenBSD 4.6 generic smp amd64 snapshot on a server > with a Realtek 8168 network card. I have problems with identiying the > root of my performance problem. I only get ~500kb/s upstream while > downstream is at maximum speed (tested via ssh and http). > > My hosting company provides me a "rescue" system with FreeBSD. It runs > on the same hardware and hasn't that problem. It gets maximum speed at > up- and downstream (the linux rescue system works as well). > > In addition this problem only occurs from the outer network of the > computing center of my server. I am in close contact with my hosting > company but we can not find any problems.
I suspect this is probably unrelated to re(4). Can you re-write "from the outer network of the computer center[...]" please? Do you mean, "from outside the network of the computer center"? Where are you testing from and what is the latency? What software are you using to test? > The only thing I found different to my testing systems are the netword > card parameters. On OpenBSD the card is always in promicous mode (and > I didn't found out how to disable it). Some software probably has set it, are you using dhclient? that would explain it. OpenBSD also negotiates flow control, I don't know if FreeBSD would list that in ifconfig output but it seems likely, in which case .Fx probably didn't negotiate it... I wonder if that might be relevant. You'd have to run a modified kernel if you wanted to disable that to test. fwiw, there's no general problem with re(4) on OpenBSD; measured by tcpbench: re sending to bge Conn: 1 Mbps: 527.317 Peak Mbps: 536.981 Avg Mbps: 527.317 bge sending to re Conn: 1 Mbps: 915.926 Peak Mbps: 929.951 Avg Mbps: 915.926
