Thanks for the detailed reply.
For clarity, the exact hardware used is:
TP-Link UE302C (RTL8156, USB-C → 2.5GbE)
UGREEN Cat6A Slim 10 Gbps, 2 m
Baseus USB 3.1 Type-C adapter
Link negotiates cleanly at 2.5 GbE full-duplex.
Tested with iperf3 using multiple streams (P4–P8), large buffers, CPU governor set to performance. MTU / jumbo frames had no measurable impact.
The result is consistently capped at ~1.7–1.8 Gbit/s. CPU usage is not the bottleneck, but all xhci-hcd IRQs remain pinned to CPU0 and IRQ affinity cannot be changed (I/O error). RPS gives only marginal improvement.
For clarity, the exact hardware used is:
TP-Link UE302C (RTL8156, USB-C → 2.5GbE)
UGREEN Cat6A Slim 10 Gbps, 2 m
Baseus USB 3.1 Type-C adapter
Link negotiates cleanly at 2.5 GbE full-duplex.
Tested with iperf3 using multiple streams (P4–P8), large buffers, CPU governor set to performance. MTU / jumbo frames had no measurable impact.
The result is consistently capped at ~1.7–1.8 Gbit/s. CPU usage is not the bottleneck, but all xhci-hcd IRQs remain pinned to CPU0 and IRQ affinity cannot be changed (I/O error). RPS gives only marginal improvement.
Statistics: Posted by dogukansahil — Thu Jan 08, 2026 6:45 pm