Am 14.02.21 um 10:35 schrieb Jaromír Doleček:
Le dim. 14 févr. 2021 à 09:27, Matthias Petermann <mp%petermann-it.de@localhost> a écrit :I would not have thought that the number of cores is so decisive here. Is the NetBSD network code SMP?I'm fairly sure it's not a CPU bottleneck - SMP or single CPU, NetBSD is fully capable of saturating 1GiB/s link. As you noted also earlier, there is plenty of available CPU. Most likely it's interrupt processing under Xen, or something along those lines.
You mean something like what is discussed here: http://www.owlnet.rice.edu/~kevinaboos/docs/interrupt-coalescing-presentation.pdf ? Kind regards Matthias