[c-nsp] 10GE sanity check

[Mikael Abrahamsson]

On Mon, 19 May 2008, Lincoln Dale wrote:

> are you operating any links like this in production today with real traffic?

I have.

> i'd love to see what (if any) CRC/Framing errors you may be recording, some 
> people think that running equipment in this manner significantly compromizes 
> the bit-error-rate of the link.
> i don't have evidence either way, but "show int x/y" counters would be a 
> great proof point of it not being an issue.

Let me give you concrete numbers that show that this is not a problem.

LX and ZX receivers are wideband (1265nm - 1600nm and 1270nm - 1600nm 
respectively).

<http://www.finisar.com/download_VoalJUFe846OFTLF1318P2xTL%20Spec%20RevA1.pdf>
<http://www.finisar.com/product-133-2_Gigabit_80km_Long-Wavelength_SFP_(FTRJ1519P1xCL)>

They do approx -3 to +5 dBm (ZX) and -10dBm to -3dBm (LX) transmitting and 
receivers are sensitive down to -22dBm (ZX, often this is -24dBm on other 
models I've seen) and -16 to -19 dBm (LX).

So let's say we have a 30km link, with 0.2dB attenuation per km at 1550, 
and 0.4dB per km at 1310. This means the ZX transmits at -3dBm and with 
30*0.2 sees 6dB loss and is received at -9 dBm. No problem for the LX 
receiver.

The LX is transmitting at -10dBm, sees 30*0.4 dB loss and is received at 
-22dBm. This is ok for the ZX receiver, but not for an LX receiver.

This works, there is no magic about it, it's just up to calculating the 
attenuation. Then again, operationally, handling this case might be 
complicated due to people having to understand the math, so I guess most 
organisations just go the easy route and get ZX at both ends as that might 
be easier as there is less understanding involved.

-- 
Mikael Abrahamsson    email: swmike at swm.pp.se