[AusNOG] Copper versus fibre in the DC

[Julien Goodwin]

On 11/10/13 21:12, Alastair Waddell wrote:
> I've been reading how copper (CAT7) is still valid with 10Gb/s ethernet
> and at the same time how the transceiver is a point of latency where the
> optics must be converted to electrical signal.

If you care about this you'd also care that the velocity factor of
copper can be faster than fibre and in which exact materials that
matters, I'm guessing you actually don't care about latency that much.

> I figure the transceiver is also a point of failure that's absent in
> copper although such an argument must surely factor the qualify of the
> cable/RJ and it's subsequent handling (but how hard can it be!)

Not quite, copper still requires line driving, it's just not a separate
cheaply replaceable module. The electronics may be simpler which was
absolutely true through fast ethernet, and I believe more of a wash from
there.

> So: 
> 
> * Is copper a valid or even a 'better' choice to terminate carriers in
> the DC for 1Gb/s and beyond to 10Gb/s? *
> 
> PS KISS and risk mitigation rule in my little world. My fallback
> position is that fibre is still preferred as the 'safe' option
> especially wrt 10Gb/s. I just want to canvass all options. I don't want
> to repeat the exercise with the carriers at some future date if I can
> avoid it. It probably means, sub 1Gb/s top-of-rack kit today (looking at
> 4948/4900M or Juniper equivalents) and new kit at somewhere near 1Gb/s
> throughput with a preference to avoid carrier re-cabling. 

There's other benefits to fibre like electrical isolation which some
people like. Also should you ever want 40 or 100g the only copper
options for the foreseeable future are those that terminate in QSFP (or
a CFP variant for 100g).

Single-mode fibre lets you cable once.


> "With the release of the IEEE 802.3an standard, 10 Gb/s over balanced
> twisted-pair cabling (10GBASE-T) is the fastest growing and is expected
> to be the most widely adopted 10GbE option. "
> 
> "At 1 Gb/s speeds, balanced twisted-pair compatible electronics offer
> better latency performance than fibre; however, considering latency at
> 10 Gb/s, currently fibre components perform better than balanced
> twisted-pair compatible 10GBASE-T electronics"
> 
> "Since optical fibre electronics cannot autonegotiate, a move from
> 1000BASE-xx to 10GBASE-xx requires a hardware change. In contrast, both
> 1GbE and 10GbE can be supported by 10GBASE-T balanced twisted-pair
> compatible equipment."

This specific case may be true (can't be bothered to research), but
there's certainly optics that can do both 100 and 1g.