RE: Equalization and benefits of Parallel Optics (skew correction)
The system Roger describes has separate clock signals. With our signalling,
each line carries its own clock information. Therefore, there are only two
limits on skew:
How far apart are we willing to put alignment characters?
How big an elasticity buffer (FIFO) will we implement?
To keep alignment simple, we want the skew to be less than half the
minimum distance between alignment characters. With a minimum distance
of 16 bytes and a 4 wide bus, we could tolerate nearly 32 ns of
skew. Increasing this much would require changing our signalling definitions
since one can currently have an A followed by a min packet followed by
another A (assuming we used the 8B/10B PCS on the link).
From: gwinn@xxxxxxxxxxx [mailto:gwinn@xxxxxxxxxxx]
Sent: Friday, July 28, 2000 3:35 PM
To: Roger Ronald
Subject: Re: Equalization and benefits of Parallel Optics (skew
At 12:19 PM 0/7/28, Roger Ronald wrote:
>It is quite possible to get rid of skew between fibers as a factor that
>distance. I speak with some authority here, since our standards committee
>(NCITS T11.1) has a 12-channel parallel 6.4 Gbit networking standard
>(HIPPI-6400) that can compensate for up to 10 nanoseconds of skew.
>This is done dynamically, with timing adjusted every 10 usecs.
>Works great. We don't care much about skew any more. Jitter is our big
>distance limiter since we have a discreet clock and our jitter budget must
>divided between two signals (clock and data).
Hmm. Ten nanoseconds maximum skew corresponds to (10*10^-9
sec)(2/3)(3*10^8 m/s)= 2.0 meters of fiber, so in a 300-meter
parallel-fiber cable, the optical lengths of all fibers in the cable would
need to agree to within 2/300= 0.7%, if fiber mismatch is allowed to take
the entire skew budget. If the link is limited to 100 meters, then the
fiber optical lengths must agree to within 2/100= 2%. These accuracies
seem a bit tight for field installations. One would probably have to use
an OTDR, and trim each fiber to fit.
How many fibers are in parallel? What's the maximum length of a HIPPI-6400
cable? What does such a HIPPI cable cost, as a ratio to say a typical
duplex fiber cable of the same length?