Matthias Petermann <mp%petermann-it.de@localhost> writes: > [ 1.000003] re0: interrupting at ioapic0 pin 18 I am not all that surprised by interrupt differences xen/not. > 1.3) iperf on localhost with Xen > > kallisto$ iperf3 -w 128K -c 127.0.0.1 > Connecting to host 127.0.0.1, port 5201 > [ 5] local 127.0.0.1 port 65534 connected to 127.0.0.1 port 5201 > [ ID] Interval Transfer Bitrate Retr Cwnd > [ 5] 0.00-1.00 sec 164 MBytes 1.37 Gbits/sec 0 492 > KBytes > > [ 5] 1.00-2.00 sec 168 MBytes 1.41 Gbits/sec 0 492 > KBytes > > [ 5] 2.00-3.00 sec 163 MBytes 1.37 Gbits/sec 0 492 KBytes > 2.3) iperf on localhost without Xen > > kallisto$ iperf3 -w 128K -c 127.0.0.1 > Connecting to host 127.0.0.1, port 5201 > [ 5] local 127.0.0.1 port 65534 connected to 127.0.0.1 port 5201 > [ ID] Interval Transfer Bitrate Retr Cwnd > [ 5] 0.00-1.00 sec 405 MBytes 3.39 Gbits/sec 0 492 > KBytes > > [ 5] 1.00-2.00 sec 511 MBytes 4.29 Gbits/sec 0 492 > KBytes > > [ 5] 2.00-3.00 sec 529 MBytes 4.44 Gbits/sec 0 492 KBytes These are quite different! The values are more or less in the same relationship as your tests using re0. But they are also far more than the most-of-1Gb/s you might reasonably expect on an actual interface. You might also try dd from /dev/zero to /dev/null. I'm seeing 16 GB/s on a sort of old box, 8 GB/s on older hardware, and maybe 3 on a PC Engines apu2. That should test the user/kernel data copying without dragging TCP into it. You might also explain your hardware and xen version. I think you said 9-stable and I'm guessing xen 4.13. I know you said "NUC5", but I wonder if the cpu dmesg lines would ring a bell for anyone. I have a box running 9-stable, that I am going to flip to xen dom0 and can try similar tests, but that's tied up with zfs issues.
Attachment:
signature.asc
Description: PGP signature