[c-nsp] Long range 10G ethernet?

[Phil Mayers]

On 16/05/12 10:18, Peter Rathlev wrote:
> (On 6500/Sup720 and with LAN cards)
>
> We're currently using a gigabit link with a total loss of 24.1dB (at
> 1510 nm) from end to end. We're using some third party "120 Km"
> transceivers, and this is working well.

Out of curiosity, is this an immensely long leg, or is it just really 
crappy? We had a link that was relatively short but had very high loss 
(though nowhere near this bad) and the fibre provider was able to 
improve this significantly more than we expected.

>
> Now we're thinking about making it a 10G link instead. Finding 10G
> transceivers capable of supporting at least ~24-25dB seems tricky
> though. Googling a bit reveals something like the Optospan
> SPP-81D-K080T31, rated for 25dB at 1310nm. Of course we're primarily
> using X2 transceivers, and that one is an SFP+.

Usually I would expect loss at 1310 to be higher than 1550, so unless 
the fibre is really odd, I'd aim for a 1550 optic on this link.

>
> Questions:
>
> 1) Any general ideas about how to best run 10G on 25+dB stretch without
> external amplification? We'd like to avoid the extra complexity that
> would introduce.

One possibility might be to use a transponder with G.709 capability, and 
appropriate optics. However, I'm not completely sure this will work - if 
your RX light level is too low, the transceiver may refuse to link-up, 
even though the extra 6dB signal/noise from the G.709 would bring it 
into spec.

*If* it works, plain ER optics should run that link in combination with 
G.709.

You didn't say what specific complexity puts you off amps, but if it's 
just "more kit" then this solution isn't really any better, and I doubt 
much cheaper, than a pair of 9dB amps.

There are also optics with embedded g.709 codecs, although I think 
they're all XFP (at the moment). This is of course simpler, but you 
can't run XFP in that platform (sigh)

> 2) We might cut ~1dB by splicing in a few places instead of patching,
> but we probably can't get below 23dB loss. Would using 25dB capable
> transceivers on a 23dB stretch be too bold for comfort?

Well... it's tight. I aim for a minimum of 3dB headroom, and that tends 
to make me uncomfortable.

> 3) Does the OneX converter support any SFP+? Or is it somehow locked to

We've got DWDM ER (40km) SFP+ in the OneX converter, and they work fine 
(both 6500/sup720 w/ 6716, and Nexus 7009 with M1 8-port).

> only support certain modules? Cisco[1] mentions only CU, SR and LR as
> supported. Any (electrical) reasons OneX couldn't support any kind of
> SFP+? Should we count on being able to use e.g. the aforementioned
> SPP-81D-K080T31 in a OneX in e.g. a 6708 or Sup2T port?

I don't see any reason this wouldn't work, absent some Cisco nonsense.