On 2026/10/19 10:31, Kevin Chadwick wrote: > > > -------- Original Message -------- > > > Instead of auto-negotiation. If you can't set both, then at least set the > > openbsd side. (by that I mean set the speed and duplex explicitly) > > then test again. > > Also if comparing Linux and OpenBSD then make sure to run the tests > immediately > after each other because StarLinks speed is surely more variable than 4G > considering sat comms fail completely in thunderstorms and both are affected > significantly by local usage rates and sat comms have even more variables > affecting speed than 4G does.
Immediately after isn't really enough to avoid this - just watch the speed of satellites moving across the sky at night. But if it's *very* low this isn't going to have much effect. (4g varies a lot too if you hit capacity on the cell or backhaul, or traffic shaping on the service). For a more reproducible test (as long as you're not in too noisy an area), testing against a local device connected to wifi would be better. I'm still interested whether there's a difference depending on cable length, which would point to different possible causes than if it's just the type of device it's connecting to - cf. the DP83815 cable length issue dealt with in a bunch of commits to sis(4).
