Options for the RS-encoded 25GE

Options for the RS-encoded 25GE
Juan-Carlos Calderon, Jean-Michel Caia
IEEE 802.3bs, May 2015, Pittsburgh, PA
Introduction
n  A PCS/FEC baseline for 25GE was adopted at the Jan 2015 Interim meeting [1]
― 
The adopted baseline defines two different formats, one without CWMs (no-FEC and CL74 FEC
modes) and one with CWMs (RS10 FEC mode)
n  An alternative approach, based on a unique format without CWMs, was proposed
at the March 2015 Plenary meeting [2]
― 
A detailed cost (and lock time) comparison of the two approaches is presented in [2]
n  In this presentation we compare 3 different options for the RS-encoded 25GE
1. 
2. 
3. 
CWMs belonging to the FEC sublayer – as per [1]
CWMs belonging to the PCS – similar to 802.3bj
No CWMs – as per [2]
n  The comparison focuses on the interaction with
a) 
b) 
c) 
1588v2
FlexE
OTN
[1] http://www.ieee802.org/3/by/public/Jan15/baden_3by_01b_0115.pdf, Jan 2015, Atlanta
[2] http://www.ieee802.org/3/by/public/Mar15/slavick_3by_01a_0315.pdf, March 2015, Berlin
IEEE 802.3by May 2015 Interim - 25G TF
2
Interaction with 1588v2 (1 of 2)
n  PTP relies on constant/known end-to-end packet delays between the Tx 1588v2
PMD
PMA
FEC
PCS
MAC / RS
1588v2
MAC / RS
PCS
FEC
Medium
PMA
PMD
1588v2
function and the Rx 1588v2 function
constant delay
n  Exact location of the 1588v2 function is an implementation choice
MAC
RS
PCS
FEC
Implementation
example 2
The delay through
FEC/PCS must be
constant
MAC
RS
1588v2
PCS
FEC
constant delay
1588v2
constant delay
Implementation
example 1
The delay through
FEC/PCS/RS/MAC
must be constant
n  A considerable amount of design effort is usually needed to minimize packet delay
variation between the 1588v2 function and the device I/O (FIFO centering, etc.)
n  A new PCS/FEC function that “moves the packets around” is not adequate for PTP
IEEE 802.3by May 2015 Interim - 25G TF
3
Interaction with 1588v2 (2 of 2)
(1) CWMs belong to FEC sublayer
MAC
Idles deleted/
inserted (gaps
for the CWMs
are created here)
RS
1588v2
packet IPG
packet
(2) CWMs belong to the PCS
packet IPG
IPG packet
PCS (enc)
Idles deleted/
inserted to insert
(Tx) and delete
(Rx) the CWMs
FEC (CWM ins/del,
Xcode, RS10)
packet delay
variation
packet
CWMs inserted/
deleted
CWM-based align
(3) No CWMs
MAC
MAC
RS
RS
1588v2
1588v2
IPG packet
constant packet
delays after this point
packet IPG
packet
IPG packet
constant packet
delays after this point
PCS (enc, CWM ins)
PCS (enc)
FEC (Xcode, RS10)
FEC (Xcode, RS10,
FEC-based align)
CWM-based align
n  The 3 different options interact with 1588v2 as follows
1. 
CWMs belonging to the FEC sublayer – as per [1]
● 
2. 
CWMs belonging to the PCS
● 
3. 
This approach is not adequate for 1588v2 – Idle insertion/deletion results on packet delay variation
No problem if the 1588v2 function (implementation) “sees” to the CWM gaps - e.g. the 1588v2 function is
disabled during the gaps
–  This is no different than what we do today for 100GE
No CWMs – as per [2]
● 
This approach is adequate for 1588v2
IEEE 802.3by May 2015 Interim - 25G TF
4
Interaction with FlexE (1 of 2)
n  A baseline text for FlexE was adopted at the 2Q15 OIF meeting in April’15 [3]
― 
The specific frame format is for further study, options under consideration are described in [3] and [4]
n  Two important concepts for FlexE (see [3] or [4]) are the use of
― 
― 
A fixed FlexE frame length
A fixed TDM structure (calendar) for the multiplexing of the Ethernet client streams into FlexE
n  The 3 different options interact with FlexE as follows
1. 
CWMs belonging to the FEC sublayer – as per [1]
● 
2. 
CWMs belonging to the PCS
● 
● 
3. 
The FEC sublayer deletes Idles belonging to the FlexE clients, corrupting the FlexE frame
No problem if the FlexE shim “sees” the CWM gaps - e.g. the FlexE shim is disabled during the CWM gaps
This is no different than what we plan to do in order to use 100GE as a server layer for FlexE
No CWMs – as per [2]
● 
No problem
[3] OIF contribution oif2015.127.01, 2Q15 OIF meeting, April 2015, Lisbon
[4] OIF contribution oif2015.139.00, 2Q15 OIF meeting, April 2015, Lisbon
IEEE 802.3by May 2015 Interim - 25G TF
5
Interaction with FlexE (2 of 2)
(1) CWMs belong to FEC sublayer
(2) CWMs belong to the PCS
MAC
MAC
RS
RS
FlexE shim
OH
1-5
...
1-5
OH
OH
Idles deleted/
inserted to insert
(Tx) and delete
(Rx) the CWMs
I
...
1-5
CWMs
inserted/
deleted
PCS (enc)
...
Idles deleted/
inserted (gaps for
the CWMs are
created here)
Idle deletion/
insertion corrupts
the FlexE frame
... CWM
OH
...
(3) No CWMs
MAC
MAC
MAC
MAC
RS
RS
RS
RS
FlexE shim
FlexE shim
1-5
OH
CWM ...
FEC (CWM ins/del,
Xcode, RS10)
FEC (Xcode, RS10)
CWM-based align
CWM-based align
FEC ...
... FEC CWM
...
1-5
PCS (enc)
PCS (enc, CWM ins)
...
OH
I
...
FEC (Xcode, RS10,
FEC-based align)
... FEC
... FEC CWM
1-5
FEC ...
FEC ...
IEEE 802.3by May 2015 Interim - 25G TF
6
Interaction with OTN (1 of 2)
n  To simplify the interconnection of FEC and non-FEC ports across OTN, ITU-T will
probably define a single format for the mapping of 25GE into ODUflex
n  As discussed in [5], the ideal case is if all 25GbE PMDs are encoded in a similar
way – i.e. if all of them have CWMs or none of them do
n  Majority of ports will be 10G/25G w/o CWMs (no-FEC & CL74 modes), therefore
― 
― 
adding CWMs to the no-FEC/CL74 modes is not an attractive option
for similarity with existing CBR adaptation functions (10GE into ODU2e, 32GFC into ODUflex), we
can assume that ITU-T Q11 will favor the mapping of 25GE into the ODUflex as a 25.78125Gbps
66b-encoded stream without CWMs
n  The 3 different options would interact with OTN as follows:
1. 
CWMs belonging to the FEC sublayer – as per [1]
● 
● 
● 
2. 
3. 
n 
OTN client adaptation function needs to convert between the CWM and no-CWM formats (cost of CWMs)
Non ideal behavior – packet delay variation
Problematic if the 25GE port carries FlexE and the OTN equipment is FlexE-unaware
CMs belonging to the PCS – same as above
No CMs – as per [2]
● 
Simplest OTN client adaptation function
● 
Best behavior (packets are not move around)
[5] http://www.ieee802.org/3/by/public/Jan15/trowbridge_3by_01_0115.pdf, Jan 2015, Atlanta
IEEE 802.3by May 2015 Interim - 25G TF
7
Interaction with OTN (2 of 2)
n  Another advantage of the approach proposed in [2] is its similarity with 32GFC
― 
― 
The main difference between an RS-encoded 25GE defined as per [2] and 32GFC is the rate
(25.78125 Gbps vs. 28.05 Gbps)
Many OTN muxponders will support multiple 25G-class clients on the same client port (e.g. 25GE,
32GFC and IB-25G), so the adaptation of an RS-encoded 25GE with no CWMs (as per [2]) would
come virtually for free
25GE / 32GFC (RS10-encoded)
Nearly identical client adaptation function
Same CBR mapping (BMP)
ODUflex
IEEE 802.3by May 2015 Interim - 25G TF
8
Summary and Conclusion
n  Based on the way the different formatting options interact with 1588v2, FlexE and
OTN, the no-CWM approach [2] is the most attractive option for an RS10encoded 25GE
CWMs belong to FEC
sublayer [1]
CWMs belong to PCS
No CWMs [2]
1588v2
Packet delays not
preserved
Packet delays preserved
(implementation dependent)
Packet delays preserved
FlexE
FlexE frame corrupted
FlexE frame not corrupted
(implementation dependent)
FlexE frame not corrupted
OTN
More complex adaptation function (cost of CWMs)
Packet delay variation
IEEE 802.3by May 2015 Interim - 25G TF
Simpler adaptation function
Packet delays preserved
25GE/32GFC reuse
9
Thank You