SJA1110EVM gPTP performance issue

[hewei5]

Hello,

We are evaluating SJA1110's gPTP performance with SJA1110-EVM. We got an issue that the time-synchronization accuracy cannot achieve to 1us as required by IEEE802.1AS in some cases (as below) which we have confirmed achievable in some other cases.

In attachment:

Sync paths.png - Two test cases which failed 1us accuracy;

CASE1_SW2P4_SW2P1_1ASResults.png and CASE2_SW2P4_P3_1ASResults.png - Test results for 2 test cases;

switch_1_config_sja1110evm.zip: Firmware used for test sample.

Test equipment: Calnex Paragon-X

What we do:

• We are trying to evaluate the gPTP performance of sync message transport paths which involve Eth ports whose speed > 100Mbps;
• We put house-made 1000Base-T1 SFP modules in each test ports (switch2-P4 and switch2-P1 in our cases);
• We test the test case 1 first and the test result is not OK. As shown in the test result of case1, Fwd CF Accuracy (avg.) is 4.958us. We are expecting it much smaller.
• For comparison purpose, we test test case 2, which has 100Base-TX port as egress port. The test result is similar, Fwd CF Accuracy (avg.) is 4.904us.

Can you please advise what could be the root cause with this issue?

In fact I have some specific concern with these test cases:

Base on gPTP principle, the gPTP protocol stack should have konwledge of the path delay from where CF parameter is born to the time reference plane of egress port, which seems to be related to the PHY who sit inside SFP module. How does SJA1110 handle this? Not sure whether this is relevent, just blaunt guess...

Thanks for help in advance!

BR,

He Wei

[hewei5]

Any idea please?

[lukaszadrapa]

Hi @hewei5 

I got a couple of questions from the team:

1. Test station – can you provide more details on that? Is that laptop with some SW, lab equipment, …? And the measurements are extracted from this device? Is there a 1PPS measurement available?

2. Is there any specific use-case which customer wants to achieve? Or just observing and testing?

3. What are the other configurations where the system is working well? Do they involve the custom SFP, or RJ-45 ports of SJA1110?

4. Could you please explain the measurements in more details? E.g., to be sure what “CF” abbreviation exactly means etc.

Regards,

Lukas

[hewei5]

Hi Lukas,

Thanks for reply. Please find below my answer inline.

1. Test station – can you provide more details on that? Is that laptop with some SW, lab equipment, …? And the measurements are extracted from this device? Is there a 1PPS measurement available?

[HW] We used a test equipment, Paragon-X, for the measurement which is specific for time error measurement. Please find in attachment its manual (CX5030 Time Aware Relay Testing_new.pdf) for more information. Test method is exactly following the instruction in the manual. As the device under test is switch (SJA1110EVM) the test doesn't requires 1PPS signal. It can be run with TX port and RX port. 

2. Is there any specific use-case which customer wants to achieve? Or just observing and testing?

[HW] The test case is depicted in attachment in previous email Sync paths.png.

3. What are the other configurations where the system is working well? Do they involve the custom SFP, or RJ-45 ports of SJA1110?

[HW] In order to run the test we used 2 house made 1000base-T1 SFP modules for Switch-2 P1 and  Switch-2 P4 respectively.

4. Could you please explain the measurements in more details? E.g., to be sure what “CF” abbreviation exactly means etc

[HW] The test precedure is exactly following Paragon-X user manual. CF = Correction Field. The value of CF field carried in sync follow-up packet of gPTP protocol.

Please let me know if any more questions. 

BR,

He Wei  

[lukaszadrapa]

Hi @hewei5 

I have additional questions to the drawing:
It obviously uses 2 SJAs on 1 EVB. These SJAs are both standalone gPTP instance? Or are they cascaded?
Then the color coded paths?. Are they representing domains? Like the red one is synchronizing the Switch2 and blue one synchronizes Switch1?
And most important question - this setup is trying to evaluate synchronization between these 2 switches? Or between test station and the EVB?

How did it go with path delay threshold update?

Regards,

Lukas

[lukaszadrapa]

Hi @hewei5 

here's next feedback:

My first advice to the customer would be to check the path delay threshold configuration.
The default value is 2us. The measured path delay above this threshold is considered incorrect. The constant offset of 5us could mean that the path delay is about 5us but the SW drops the path delay measurement result?and does not compensate the time with path delay at all.

Please advise the customer to update threshold to 10000 (ns) and run the tests again.

Regards,

Lukas

[lukaszadrapa]

I'm sorry for delay, I had to forward this to our experts, I'm not able to answer this directly. Thanks for patience.

Regards,

Lukas