(I admit I had no idea what RMII was. See: Media-independent interface - Wikipedia )
Thanks, @DrYak . I was definitely well off the mark, there. I checked with iperf3
, and got 800Mbps-900Mbps off the Pi. Average was around 867 Mbps or so, which is what I would think would be normal given overhead? I’m not using Jumbo Frames or any other optimization, as I’d need to understand not only how to do the optimization, but figure out what other devices need to also have those optimizations done and how to isolate them on a VLAN to avoid messing with the slower devices on the network. I’ve never even set up a VLAN before, so…one thing at at time. 
(I saved the results and may try to figure out how to make a fun graph out of them.)
I love Jeff Gearling’s stuff, btw. Watching him got me interested in pushing my Pi further, even if I don’t really have a great need to do so.
I didn’t realize the USB3 connection could work at up to 4 Gbps.
I think the CPU is definitely a bottleneck. I’ve got a USB 3.2Gen2x2 (?!) enclosure for an MVME drive that I’m using as my boot drive. … Okay, it’s 20Gbps, whatever that flavor of USB is this week. Certainly fast enough not to come anywhere close to being a bottleneck on a Pi.
On a read test using hdparm
, I see ~300MB/sec, which from Jeff Gearling’s results is exactly what I should be seeing on a drive going as fast as it can on that bus. That’s 2.4 Gbps on a bus with a theoretical max speed of 5 Gbps.
Is that 2.4 Gbps because (1) each USB 3.0 port gets half the available 5 Gbps of the USB 3 bus, shared between them; or (2) the CPU itself really is that much of a bottleneck?
If it’s (1), I’d expect good results with the 2.5GbE adapter. Anything in the 1.8* Gbps-2.2Gbps range would seem to indicate it’s working well.
- Given my
iperf3
result with the onboard Ethernet, I’d expect pair bonding two 1Gbps ports to hit about 1.6-1.8 Gbps. I’d expect a single 2.5Gbps port to outdo a pair of slower ones, unless it really does choke the CPU.