Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

RE: [802.3ae] Regarding Sequence ordered sets (Clause 48)




Sanjeev,

That behavior would not cause any harm. It would delay sending the RF
ordered set by one column, but that is a negligible effect since the sending
of RF is not time critical. I believe that the contents of 46.3.3.2 make it
an allowed behavior. 

It is not a required behavior. 46.3.3.1 makes it clear that a frame ended by
a control character other than Terminate causes the same error as receiving
an Error control character during a frame. Other physical sublayers that
start to send local fault do not insert an error character before doing so.
Therefore, the RS must react to both behaviors by ensuring that the frame is
discarded.

Regards,
Pat

-----Original Message-----
From: Sanjeev Mahalawat [mailto:sanjeev@cisco.com]
Sent: Wednesday, November 28, 2001 4:42 PM
To: pat_thaler@agilent.com; bob.grow@intel.com; stds-802-3-hssg@ieee.org
Subject: RE: [802.3ae] Regarding Sequence ordered sets (Clause 48)


Hi Pat,

RS may choose to truncate the ongoing frame with inserting /E/ in frame and
delineating it with /T/.
I would think this should be legal to do too rather than just transitioning
from Data to Idle or
sequence_ordered sets. In anycase it could be fragment and neither case
packet will be accepted.
Do you see any harm in it?

Thanks,
Sanjeev

At 02:56 PM 11/28/2001 -0700, pat_thaler@agilent.com wrote:
>
>Bob,
> 
>Good explanation.  I want to clarify something. Naresh asked: "Is it
>possible for sequence ordered sets to appear in the middle of a data
frame?"
> 
>The answer is sequence ordered sets cannot appear "in the middle of a data
>frame" One will not see Start of a frame/sequence ordered sets/remainder of
>that frame. 
> 
>When the RS is sending a frame and needs to switch to sending a sequence
>ordered sets, that ends the frame in an errored state. One wouldn't want
the
>RS to put a T character out to terminate the frame because that might
create
>a fragment that looks like a valid frame (if by chance the last 4 bytes had
>a correct CRC for the prior content). The rules in clause 46 cover
receiving
>a transition from data to control without a T.
>  
>An intermediate device such as a PCS may detect a fault condition and
>similarly transition from sending data to sending the Local Fault sequence
>ordered set.
> 
>Also, note that at the receiver, the transition from data to sequence
>ordered sets may appear as a transition from data to idle (without a T) if
>there is an XGXS or a 10GBASE-X PCS between the transmitter and the
>receiver. These sublayers convert a series of sequence ordered sets into an
>idle with sequence ordered sets interspersed every 16 to 32 columns to
>prevent the EMI effects of sending a repeating data pattern.
> 
>Pat
> 
>-----Original Message-----
>From: Grow, Bob [mailto:bob.grow@intel.com]
>Sent: Monday, November 26, 2001 4:49 PM
>To: 'stds-802-3-hssg@ieee.org'
>Subject: RE: [802.3ae] Regarding Sequence ordered sets (Clause 48)
>
>
>I don't what your clause 48 concern is.  I think the possible cases are all
>described in clause 46.  
> 
>A Remote Fault sequence ordered set is only generated by the RS, and only
in
>response to detecting a Local Fault condition on the receive path.
>Conceptually, the link fault state machine on the receive path switches a
>transmit path mux from MAC frames to RS generated Remote Fault ordered
sets.
>This is done without sending a Terminate (sending a Terminate would
increase
>the undetected error rate).  Action is similar for detection of a Remote
>Fault condition, with the exception that a frame in transmission would be
>truncated with Idle instead of sequence ordered sets.  
> 
>Transmission of the Remote Fault sequence ordered sets (or Idle) continues
>until the link fault condition is cleared.  It is theoretically possible
>when the condition is cleared to be in the middle of a transmitted frame
>(either the same frame or a different one.  At the remote receive RS, the
>reception of either Idle or Remote Fault ordered sets will terminate the
>initial starting frame fragment as an error frame.  The trailing frame
>fragment (no Start) will be ignored by the remote receiving RS.
> 
>So, only looking at the data stream, at the receiving RS, a beginning frame
>fragment and an trailing fragment can be separated by a string of columns
>starting with a non-data column.  The standard only describes generation of
>that string as Local Fault sequence ordered sets, Remote Fault sequence
>ordered sets, or Idle.  The receiving RS though will behave the same if
>something else is received within the string (e.g., a corrupted sequence
>ordered set, inserted Idle, ect.) 
> 
>Therefore, a PCS encoder or decoder can will be hit with changes from data
>to ordered sets, data to Idle, Idle to ordered sets, ordered set to
>different ordered set, etc.
> 
>--Bob Grow
>
>-----Original Message-----
>From: Naresh Raman [mailto:naresh@lsil.com]
>Sent: Sunday, November 25, 2001 8:36 PM
>To: stds-802-3-hssg@ieee.org
>Subject: [802.3ae] Regarding Sequence ordered sets (Clause 48)
>
>
>All,
> 
>    I have a question regarding Sequence ordered sets. Is it possible for
>sequence ordered sets to appear in the middle of a data frame?
> 
>    Clause 46.3.4 (lines 27-29) says that the RS can possibly truncate a
MAC
>frame being transmitted to generate Remote Fault Status over the transmit
>path. I would like to know if this means that there will be a sequence
>ordered set inserted in the middle of the MAC frame or will the MAC frame
be
>terminated (by sending a Terminate control character) before the Sequence
>ordered set is inserted in the transmit path.
> 
>Thanks,
>Naresh Raman.
>LSI Logic.
>