Xilinx has a report on Infiniband cable
characterisation with RocketIO
MGTs (v2.0, May 10th 2004,
http://direct.xilinx.com/bvdocs/reports/ug043.pdf). Unfortunately,
this is vague on BER measurements. The intro states that BERs are
presented for various configurations but, in the event, the data
summary table only states whether or not the configuration "meets RX
requirements", which presumably means whether or not the BER is <=
1E-12. The tests also used PRBS 2E7 rather than 8B/10B encoding, so
the BERs wouldn't be correct for Infiniband anyway, or for Aurora.

Anyone happen to know of any better figures, or a later report? Or
have any experience of measuring BERs on RocketIO over these cables?

Thanks -

Paul

Paul Johnson wrote:
> Xilinx has a report on Infiniband cable characterisation with RocketIO
> MGTs (v2.0, May 10th 2004,
> http://direct.xilinx.com/bvdocs/reports/ug043.pdf). Unfortunately,
> this is vague on BER measurements. The intro states that BERs are
> presented for various configurations but, in the event, the data
> summary table only states whether or not the configuration "meets RX
> requirements", which presumably means whether or not the BER is <=
> 1E-12. The tests also used PRBS 2E7 rather than 8B/10B encoding, so
> the BERs wouldn't be correct for Infiniband anyway, or for Aurora.
>
> Anyone happen to know of any better figures, or a later report? Or
> have any experience of measuring BERs on RocketIO over these cables?
>
> Thanks -
>
> Paul

BER over IB cables have a specific compliance measurement requirement,
which I unfortunately do not have access to at the office, but I'll
look it up later (I was a member of the physical layer subgroups of
both EWG and CIWG amongst other things and I have the docs at home).
I would assume Xilinx used that compliance spec requirement for the
test (if they didn't, they obviously should not be claiming
compliance).

The specific requirements for IB cable (for single data rate, 2.5Gb/s
per pair) is

BER : <= 1E-12
Attenuation: <=10dB

For 4x and 12x cables, (again SDR) there are also specific xtalk
requirements.

There are also jitter spec requirements which I can look up once I get
home.

Note the requirements for DDR and QDR cables are somewhat different, as
the loss budget was changed from a loss style measurement to an
eye-closure method of measurement for the system, and the compliance
requirements were changed to S-parameter form.

Can't help on RocketIO, though.

Cheers

PeteS