I am seeing a repeatable, receive-specific performance problem when an external Plugable USBC-E2500 (Realtek RTL8156B) 2.5GbE adapter is connected through my Plugable TBT4-UD5. The same adapter performs substantially better when connected directly to the laptop.
Why am I doing this in the first place? The TBT4-UD5 only has a 1GbE built-in NIC and my home office has a 2.5GbE path to my basement server rack.
System:
• Laptop: Surface Laptop 5
• Dock: Plugable TBT4-UD5 connected to laptop USB-C
• Network adapter: Plugable USBC-E2500 (Realtek Gaming USB 2.5GbE Family Controller, RTL8156B)
• Realtek driver: 1156.23.20.605
• Ethernet link: 2.5 Gbps
• Test server: QNAP NAS running iperf3
• SMB Multichannel disabled
The network path has a separate MoCA-related receive ceiling of approximately 440 to 540 Mbps. I am not expecting full 2.5GbE receive throughput across this path. The issue reported here is the additional reduction to approximately 140 to 160 Mbps that occurs only when the adapter is connected through the TBT4-UD5 dock.
A/B results using the same laptop, adapter, Ethernet cable, QNAP, and network path:
Adapter connected directly to laptop USB-A, with dock still attached to laptop USB-C:
• Laptop to QNAP TCP: 2.21 Gbps
• QNAP to laptop TCP: 440 Mbps
• QNAP to laptop, four streams: 633 Mbps
• SMB read: 53.55 MiB/s
• Durable SMB write: 137.83 MiB/s
Adapter connected to the dock’s rear 10Gbps USB-C port:
• Laptop to QNAP TCP: 2.29 to 2.32 Gbps
• QNAP to laptop TCP: 136 to 148 Mbps
• QNAP to laptop, four streams: 227 to 233 Mbps
• SMB read: 15.57 to 16.81 MiB/s
• Durable SMB write: 142.08 MiB/s
Adapter connected to the dock’s front downstream Thunderbolt port:
• Laptop to QNAP TCP: 2.17 Gbps
• QNAP to laptop TCP: 159 Mbps
• QNAP to laptop, four streams: 198 Mbps
• SMB read: 17.96 MiB/s
Topology:
Windows USB topology showed that the rear port placed the adapter behind a VIA Labs hub:
• USB\VID_2109&PID_0822
The front Thunderbolt port instead placed it behind an Intel SuperSpeed hub:
• USB\VID_8087&PID_0B40
Bypassing the VIA hub therefore did not appear to correct the problem.
Additional controls:
• The dock’s built-in 1GbE adapter reached 470 Mbps in the same receive direction with iPerf3, and approximately 49.44 MiB/s for SMB reads, outperforming the 2.5GbE adapter by 3X.
• The external adapter negotiated at 2.5 Gbps in every configuration.
• Windows reported zero received errors, received discards, outbound errors, or outbound discards.
• Disabling Receive Segment Coalescing did not help, so it was restored.
• Forward TCP and SMB writes remain fast through the dock. The severe degradation is specifically in traffic received by the laptop.
• Keeping the dock connected while moving only the adapter to laptop USB-A restores the expected performance. The dock’s general presence and display workload are therefore not sufficient to cause the slowdown.
Current workaround:
I am keeping the TBT4-UD5 connected to the laptop’s USB-C port for power, display, and peripherals, while connecting the 2.5GbE adapter directly to the laptop’s USB-A port.
The remaining gap appears to be an interaction between the RTL8156B adapter and the dock’s downstream USB paths, rather than a simple link negotiation, Ethernet cable, VIA-hub-only, or overall dock bandwidth problem.
Could Plugable advise whether there are known RTL8156B compatibility issues, relevant dock or controller firmware updates, or additional diagnostics that would help isolate this? I emailed a PlugDebug capture in and can provide more detail if useful.