Top ten reasons to love (and hate) 100 Gigabit Ethernet
Before I list the reasons to be excited by the prospect of implementing 100 Gigabit Ethernet and developing its components, let me lay out its essential aspects.
There are two topologies: 10×10 and 4×25; 10×10 has ten separate 10 Gb/s lanes and 4×25 has four at 25 Gb/s. By “10 Gb/s” and “25 Gb/s” I’m referring to the payload. The actual transmission rates are 10.3125 Gb/s and 25.78125 Gb/s, respectively – the excess accounts for the two bits of overhead per 64 bits of data in the 64B/66B encoding scheme.
- 100GBASE-CR10 = 10 x 10 on shielded balanced copper cabling with reach up to 7 m.
- 100GBASE-SR10 = 10 x 10 on ten separate multimode fibers with reach up to 100 m.
- 100GBASE-LR4 = 4 x 25 multiplexed (WDM) on one single-mode fiber with reach of at least 10 km.
- 100GBASE-ER4 = 4 x 25 multiplexed (WDM) on one single-mode fiber with reach of at least 40 km.
As yet there is no backplane specification for 100 GbE, though there is for the 4×10 flavor of 40 GbE.
The electrical aspects of the physical layer specification leverage 10 GbE for each of the 10 Gb/s lanes but is cagey about the 25 Gb/s electrical signaling that will obviously be needed. Don’t worry though, until the 4×25 electrical signaling is specified you can leverage the Optical Internetworking Forum’s work on the common electrical i/o (OIF-CEI) — but feel free to tinker.
In any case, like other high speed serial (HSS) technologies, 100 GbE electrical channels use differential signaling with embedded clocks and de-emphasis is prescribed at the transmitter and equalization at the receiver. The parallel 100 GbE configurations are essentially independent HSS channels with all jitter advantages and clock recovery hassles included at no extra cost! In this sense, 100 GbE is nearly a complete superset of the hassles you face with the 2nd, 3rd, … generations of the other HSS technologies.
Here’s my top ten list of 100 GbE intriguing features:
9. Until the 4×25 electrical signaling is specified you have freedom to tinker and/or can leverage from the Optical Internetworking Forum’s work on the common electrical i/o (OIF-CEI)
8. The physical coding sublayer (PCS) “gearbox” shuffles the issue of interlane skew higher in the protocol stack, so we don’t have to worry about it at the physical layer.
7. Forward error correction (FEC) is optional (so far).
6. Most of the optical issues are well trod, but clever ways to multiplex signals onto single fibers, like DP-QPSK are emerging. Using quadrature phase shift keying with optical signals is one thing, but doubling the data rate by sending another pair of signals on the same fiber with different polarizations is genius.
5. New jargon for crosstalk! Victims will hereby be annoyed by disturbers rather than disturbed by aggressors.
4. Since each lane can use an independently recovered clock, there is no requirement that separate lanes be synchronized which will make diagnosing crosstalk problems considerably more interesting.
3. The integrated crosstalk noise (ICN) requirement combines limits on insertion loss (IL) and crosstalk noise in an ICN vs IL template – we love templates!
2. Flexible and delightfully complicated interference tolerance testing: two tests that probe the receiver’s ability to tolerate crosstalk in the presence of jitter. One with low crosstalk and maximum channel insertion loss, the other with maximum crosstalk and minimum insertion loss.
1. Transmitter de-emphasis amplifies high frequency signal content to compensate the low pass nature of the channels, but those big voltage swings will generate much worse crosstalk.
0. Equalization in the presence of crosstalk is rife with opportunity. While continuous time linear equalization (CTLE) and feed forward equalization (FFE) both amplify crosstalk noise, you can just smell different ways to equalize away crosstalk. They even give you a training sequence during initialization.
It’s rife, I tell you!