[8023-10GEPON] Should we use shortened FEC frames
Dear colleagues,
I have updated my previously posted spreadsheet calculating FEC overhead
with and without shortened last FEC frame.
Here is a brief summary of the "shortened last FEC frame" feature. Consider
a situation when an ONU needs to send one 1000-byte packet.
If we use (as an example only) RS(239, 255), we would send 239 bytes in the
first FEC frame followed by 16 bytes of parity, then again 239 bytes
followed by 16 bytes of parity. After 4 FEC frames, we will have 1000 -
4*239 = 44 bytes of data left.
Here, if we don't allow a shortened block, the ONU will have to send an
entire FEC frame: 239 bytes of payload + 16 bytes of parity. The payload
will consist of 44 bytes of data plus padding.
If we allow shortened block, the payload would consist of 44 bytes of data
only, followed by 16 bytes of parity.
What is the catch? There are a few actually:
a) Complexity: Delineation of payload and parity is more complicated when
payload can be of variable length. That would require more complicated
circuitry at both ends.
b) Delineation: When all FEC frames are of the same length, it is easy to
delineate payload from parity simply by their position within FEC frame.
With variable payload size, special delimiters are required. 1G EPON uses 3
delimiters per frame (S_FEC, T_FEC_x, T_FEC_E). These delimiters are either
5, 6, or 7 bytes long and rely of special characters.
But with 64b/66b coding, any 66-bit value may be only 2 bits away from a
valid data block. It is not clear how to delineate variable payload from
parity.
c) Overhead: Assuming that some delimiters can be found, what is more
efficient: (a) to use these additional delimiters in every frame or (b) not
to use a shortened frame at all?
This updated version of the spreadsheet takes into account additional
delimiters needed for shortened FEC frames.
The calculation shows that with only 8-byte delimiters per FEC frame, it is
better not to use shortened FEC frames.
For example, without shortened frames, FEC overhead is 6.81%, while with
shortened frames it is 9.50% (see excel file for this or other setups).
I encourage you to do the following before the next meeting:
1) Open the attached excel file and play with different numbers. You may
easily add additional columns and try additional setups.
2) Let me know if you disagree with any of the calculations.
3) If you still argue in favor of shortened blocks, please make a detailed
proposal regarding delineation.
I plan to make a few slides for the next meeting outlining my conclusions
and I would appreciate your feedback now, so I can include it in my
presentation.
Thank you,
Glen
FEC_overhead_Ver4.xls