[802.3_ISAAC] Discussion on Delay Constraints & Annex 191A - Comments #77 and #78

[Ragnar Jonsson <ragnar.jonsson@xxxxxxxxx>]

All

 

I have concerns about comments #77 and #78 related to Delay Constraints & Annex 191A

 

Background

• Comments #77 and #78 observe that end-to-end delay constraints in 191.14 are difficult to enforce across two different vendors. They propose replacing the end-to-end constraint with normative limits on individual Tx and Rx delays, which would effectively change Annex 191A from Informative to Normative.

 

1. PAUSE Utility in Camera Links

The primary justification for splitting the delay constraints in the draft is to ensure the "predictable operation of the MAC Control PAUSE operation."

However, considering that the primary target applications for this asymmetric PHY are camera and sensor links, MAC Control PAUSE has a very limited practical purpose. The Task Force should review how PAUSE would be utilized in these specific topologies, and whether that use case justifies placing additional constraints on PHY optimization.

 

2. The Informative Nature of the Annex 191A Split

The specific percentage splits between the Tx and Rx delay paths in Annex 191A were originally developed and accepted by the Task Force as informative guidelines.

Making these allocations strictly normative may unnecessarily reduce PHY design flexibility. For example, a vendor might design a highly optimized Tx path alongside an Rx path that is slightly slower than the exact Annex 191A allocation. If their combined delay is acceptable for the system, they should ideally remain compliant.

Furthermore, if the delay allocations are to be strictly enforced as normative, the current values may need to be re-evaluated. For instance, it would likely be more practical to use a 15% / 85% split instead of the current 10% / 90% split for the HS_PATH.

 

3. Proposed Paths Forward

If the Task Force determines that the delay constraints should be modified, I suggest two potential approaches:

 

Option A: The System Integrator Assumption

• Keep Annex 191A Informative.
• Provide guidance that if system integrators suspect that the specific PHY combination they are using may violate the end to end delay, they can simply assume a conservative XGMII-to-XGMII delay (for example, 10 µs for the LS_PATH) for the purpose of calculating their PAUSE buffers.
• This provides integrators with a safe ceiling for system calculations while maintaining design flexibility for PHY implementers.

 

Option B: Relaxing the Limits

• If the Task Force decides there is true value in normatively restricting the delay separately in the transmit and receive paths, the overall limits would likely need to be relaxed to provide sufficient margin, particularly on the LS_PATH.
• For example: Increase the overall LS_PATH XGMII-to-XGMII latency limit from 5 µs to 10 µs, and apply the agreed-upon percentage split to that larger budget.

 

4. Consistency with Clause 192

Finally, the Task Force should consider how these delay requirements apply to Clause 192. It would be beneficial to ensure that our approach to PAUSE timing and delay budgeting remains consistent across both asymmetric PHY clauses.

 

 

Ragnar

 

---

To unsubscribe from the STDS-802-3-ISAAC list, click the following link: https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-ISAAC&A=1