QSW-M408S with an RJ45 SFP+ module: 2.5GbE link crawls at 140 Mbit/s while plain 1G is fine
Homelab, nothing exotic. An Odroid H2 does routing duty on its 2.5GbE ports, and I wanted to hang it off one of the 10GbE SFP+ cages of the switch through an RJ45 copper module instead of pulling fibre across the room.
- QNAP QSW-M408S, copper module in one of the 10GbE SFP+ cages
- Odroid H2 router, onboard 2.5GbE ports
- RJ45 (10GBase-T) SFP+ transceivers, one 6Com and one GWfiber
- short Cat6a patch cable, both boxes sitting on the same bench
The link comes up and passes traffic, but the throughput is a joke:
2.5GbE port -> RJ45 SFP+ module -> QSW-M408S SFP+ cage
iperf3: about 140 Mbit/s
same router, plain 1GbE RJ45 port on the switch
iperf3: full gigabit, no complaintsWhat I tried:
- swapped between the two modules, the 6Com and the GWfiber behave identically
- changed the patch cable and moved the module to a different SFP+ cage
- fell back to the plain gigabit RJ45 port to prove the router and the cabling are fine, which is where the gigabit line above comes from
So is the switch simply not doing 2.5G through a copper SFP+ cage, or are these modules the wrong thing for multigig?
Also available in: English
العربية
Deutsch
Español
Français
हिन्दी
Bahasa Indonesia
Italiano
日本語
한국어
Nederlands
Polski
Português
Русский
Türkçe
Tiếng Việt
简体中文
Comments 5
The switch is not what limits you here, the modules are.
Copper SFP+ at the cheap end typically negotiates 1G and 10G and nothing else. The intermediate multigig rates are a separate feature: a module that supports them says so, listing 2.5G and 5G alongside 1G and 10G, because that is NBASE-T rather than plain 10GBase-T. When the module cannot do the rate the other end wants, you get exactly your symptom, a link that comes up, technically works and moves a fraction of what it should, while the plain gigabit path is perfect because 1G is one of the rates the module does handle.
So before blaming the QSW-M408S, get a module that explicitly lists 2.5G and 5G in its specification. Wiitek is the one that came up when I was looking at the same thing. If the datasheet stays silent about those rates, assume the module cannot do them, whatever the seller writes in the listing.
I have not run that combination on a QSW-M408S myself, so buy one module and test it before ordering a handful.
What do those two modules actually claim? Not the marketing on the listing, the rate list in the datasheet. Plenty of the cheap 10GBase-T SFP+ only ever advertise 1G and 10G, and everything in between is left to fall where it may.
Also, does the switch tell you which rate the port settled on, or only that the link is up? And what does the H2 report on its side? A port that thinks it is at 2.5G while moving 140 Mbit/s and a port that quietly dropped to something lower are two different problems with two different fixes.
Same story on different hardware. I had a Chelsio T520-CR and asked around about copper SFP+ before buying: the answer was that FS-branded RJ45 modules run in Chelsio cards at 1G and 10G with no special handling, and people had been running the same modules across Chelsio and Intel NICs for a long time without trouble. The moment the conversation turned to 2.5G and 5G, everyone got vague. That is the part copper SFP+ is unreliable at, in that discussion as much as in yours.
The advice I ended up following was to stop buying transceivers and get a T520-BT with native RJ45 ports instead. If the box you are connecting has usable native copper ports, the SFP+ cage is better spent on fibre or a DAC.
Worth adding to the shopping list: power budget, not only the rate list.
A colleague ran a Mac Studio into an SFP+ RJ45 module in a USW Aggregation over a 50-foot Cat6a run, with a 10GbE NAS on another cage. Autoneg never got him to 10G there, the port settled on 5000Base-T every time, and pinning 10GBASE-T by hand made it worse rather than better: copies dropped to somewhere around 100 MB/s. The conclusion was that adapters like that run out of power budget and only hold 10G on short runs. What did help: fixing the rate by hand so the power-saving logic stays out of it, switching flow control on, and shopping for a module with more power behind it, the MikroTik S+RJ10 being the one named. The alternative was dropping copper modules altogether and moving to a switch that has 10GbE RJ45 ports of its own, a USW-Flex XG in that case, leaving the cage for fibre or DAC.
On a bench patch cable none of this shows up, but it matters as soon as the cable goes through a wall.
Half agree. The modules are probably too limited, but I would not order anything before looking at how the port negotiates.
I have a CRS305-1G-4S+IN on SwOS 2.18 with a 1 Gbit-only 10Gtek RJ45 SFP sitting in an SFP+ cage. It runs cool and has been solid, but only because auto-negotiation is switched off on that port and the rate is pinned to 1G. Leave autoneg on and an SFP+ cage talking to a module with a narrower rate list never settles properly, which is the same family of problem you are describing.
If the QNAP lets you fix the port rate, pin it to 2.5G on both ends and repeat the iperf3 run. It costs nothing, and it tells you whether you are buying a new module or chasing negotiation.