Here's a list of the five biggest challenges facing the new generation of 25G Ethernet products in hopes they will be addressed sooner rather than later.
The UNH-IOL has been testing Ethernet products since its inception in 1988, and over the years we have come across some Ethernet products that really puzzled us. Here are the top problems we see.
Not supporting or disabling auto-negotiation: The importance of auto-negotiation (Aneg) to 100G interoperability was acknowledged by making it mandatory to be fully compliant to IEEE 802.3 Clauses 92 and 93. The more recent 25G standard also mandates Aneg.
Not supporting Aneg could result in both links not resolving at the maximum mutually supported rate or the inability to link at all. Aneg also serves as the critical precursor step to training which optimizes the electrical signaling to offset the negative effects of the channel.
Two different, yet similar specifications: The 25G/50G Consortium’s work put IEEE 802.3by on the fast track for standardization. Thanks to the 25G/50G Consortium, there were 25G capable devices before the IEEE standard was complete.
However, just because a solution is compliant to one does not make it compliant to the other. In fact, it is likely that a device won't initially be compliant to both. While the electrical conformance specifications may be similar, the slight differences in speed negotiation and forward error correction can cause confusion that requires fluency in both technologies.
Inconsistent use and understanding of cable/module information: EEPROM information located on modules and cables was supposed to help a device understand basic channel properties such as length, loss or module type. Some manufacturers have leveraged that information to lock out other manufactured products. Others have misinterpreted the data.
For example if the cable advertises it has the physical capabilities to handle 25G rates, but for some reason a device-under-test wants to do 40G (10G rate), the device may lock down because it didn’t have its logic correct, even though it could definitely operate over the slower rate than advertised on the cable.
Understanding of the specification and compliance: Due to the nature of the technology, the IEEE 802.3 specification for 25G is a dense and complex document.
Take, for example, the Channel Operation Margin (COM) procedures and scripts. While it is fairly straightforward to perform the necessary testing, understanding how changes need to be implemented based on the results is not. This is really just one of many interpretation issues that are inherent in such a standard.
In addition to the interpretation issues, many devices are not fully compliant to the specification because they were designed while the specification was being drafted. These pre-standard devices complicate testing and interoperability as they do not share the plug-and-play pieces that make Ethernet so successful.
Diverse applications and marketing confusion: This is a great problem to have. There are many emerging applications for 25 Gigabit Ethernet, from the hyper-scale data center to cloud applications and campus deployments. This is a natural evolution of the Ethernet roadmap and shows a commitment to break with old paradigms that are no longer relevant, such as always increasing the data rate by 10 times.
Another paradigm that is changing is the traditional one module or cable per port. Many devices are taking advantage of the flexibility offered by both the single lane nature of the 25 Gb/s specification as well as MSA port types such as QSFP28 to connect multiple servers in a rack to a single top or middle of rack switch. This flexibility allows many different applications to take advantage of the standard and maximizes the bandwidth available in switches.
--Jeffrey Lapak is the Enterprise Industry and Operations Strategic Manager and Associate Director at the University of New Hampshire Interoperability Laboratory. He manages all Ethernet related testing and test events.