I have two servers, the first with a built-in and add-on 10gbit port, and the second with built-in gbit and add-on 10gbit (Syba AQC107), communicating through a feed-through rated for gigabit communication (Cat6A shielded cabling totaling ~50ft, unshielded couplers). Thus, I don't expect full 10gbit communication to work. However, the two built-in ports communicate reliably and I get full gigabit transfer speeds. When one of the Syba ports is connected at either end, the autonegotiation is unreliable, and usually falls back to 100Mbit. However, when it does succeed at gbit autonegotiation, it does get full gigabit transfer speeds. When both Syba cards are plugged in, no link is detected at all. Connecting the Syba cards directly (50ft cable, no feedthrough) results in 10gbit throughput.
I infer from this that the current constraint is the Syba port's autonegotiation capability, and not the physical cabling. Thus, I hope that I may be able to at least get 2.5Gbit transfer speeds with a different NIC, but I don't have any way to know a priori that a given card will be suitable without the impedance, line filtering, etc. of the port. Is this simply an instance of money=power? Does anyone have brand suggestions that might prove fruitful?
Related, how is it possible that these cards can reliably get gigabit throughput, but frequently (~80% of time) fail to autonegotiate gigabit speeds? Is autonegotiation typically handled by a different DSP?
I am using ethtool for NIC status and iperf for transfer rates.