SJA1110 Qbv Cycle time min value

[hewei5]

Hello,

I am evaluating SJA1110 Qbv configuration with SJA1110EVM.

When the cycle time of gate control list is 1ms (125us * 8 slots) everything is fine.

When the cycle time is decreased by 10, etc. 100us = 12.5us * 8 slots, it seems some packets are being blocked.

Could you please advise the lower limit of cycle time and slot time for  the gate control list of SJA1110?

Thanks!

BR,

He Wei

[hewei5]

Hello Abdullah,

The setting has been modified as instruction. Please find it in attachment. While it seems not working :-(.

Could you please help confirm the setting?

Thanks a lot!

BR,

He Wei

[abdullahzafar]

Hi Hewei5, 

Yes, the configuration seems okay to me. Did you manage to read the drop counters? 

DIAG_CNTRS and other drop counters. This will help to indicate on why the frames are 

getting dropped. 

BR, 

Abdullah 

[hewei5]

Dear Abdullah,

Thanks for response!

We have modified the configuration accordingly as attachment and rerun the testing.

While it seems not working...

Would you please help to double check and confirm the setting is modified as per your instruction?

Really appreciate!

BR,

He Wei

[davidtosenovjan]

I am forwarding an expert's answer:

"We dont have a defined lower limit for the Admin cycle time. However, "packets being blocked" is generally
when sum of trigger time > admin cycle time. The first step should be to ensure that admin cycle time > trigger time.
Moreover, how exactly is the customer configuring the switch, is it using our SDK? and are they using PE micro or Lauterbach
for switch configuration? The reason why I am asking this is if admin cycle time > trigger time, and even then the packets are
being dropped. then it makes sense to read the drop counters. This will give a better indication on why packets are being dropped."

[hewei5]

Hello, Thanks for reply!

We are trying to enable Qbv on a path with mixed bandwidth, from SW2-P1（1000Base-T1 SFP） through SW1 - P1（100Base-TX）。

The GCL setting is as following. The idea is to create cycle time 1ms = 125us /slot * 8 slots.

For  SW2-P1(1000Base-T1):

Schedule Table 0, Gate Status 0~7:10110001 Trigger Time:625
Schedule Table 1, Gate Status 0~7:10110001 Trigger Time:625
Schedule Table 2, Gate Status 0~7:10110001 Trigger Time:625
Schedule Table 3, Gate Status 0~7:10110001 Trigger Time:625
Schedule Table 4, Gate Status 0~7:10110001 Trigger Time:625
Schedule Table 5, Gate Status 0~7:11001111 Trigger Time:625
Schedule Table 6, Gate Status 0~7:11001111 Trigger Time:625
Schedule Table 7, Gate Status 0~7:11001111 Trigger Time:625

QBV Administrative Parameters for Port1：
ADMINCYCLETIMEEXT:7813
ADMINCYCLETIME:62500
ADMINGATESTATES: GATE 0~7: 11111111
QMAXSDU[0~7]:2047

For  SW1 - P1（100Base-TX）:

Schedule Table 0, Gate Status 0~7:10110001 Trigger Time:63
Schedule Table 1, Gate Status 0~7:10110001 Trigger Time:63
Schedule Table 2, Gate Status 0~7:10110001 Trigger Time:63
Schedule Table 3, Gate Status 0~7:10110001 Trigger Time:63
Schedule Table 4, Gate Status 0~7:10110001 Trigger Time:62
Schedule Table 5, Gate Status 0~7:11001111 Trigger Time:62
Schedule Table 6, Gate Status 0~7:11001111 Trigger Time:62
Schedule Table 7, Gate Status 0~7:11001111 Trigger Time:62

QBV Administrative Parameters for Port1（100Mbps）：
ADMINCYCLETIMEEXT:7813
ADMINCYCLETIME:62500
ADMINGATESTATES: GATE 0~7: 11111111
QMAXSDU[0~7]:2047

Would you please advise if it is correct?

We are using SDK and PE micro for the setting. 

Going through software user manual there seem to be quite some registers related to the package dropping. Could you please suggest which registers we could start with to look into a bit deeper for our issue?

Thanks!

BR,

He Wei

[davidtosenovjan]

I am forwarding an expert's answer:

1.
The admin cycle time needs to be 125000 if you want it to be 1 ms. 62500 seems wrong.
Wrong example: admin cycle time of: 250000000 equals to 2 seconds. please adjust accordingly.

2.
If this still does not work, please share the project directory with me.

3.
the ingress traffic and the egress port. would be nice if its a block diagram.

[hewei5]

Hi,

Highly appreciate if you can help reviewing the config.

Please find in attachment our settings:

Qbv data path.png - The block diagram with traffic path 

the other two .zip files -  the settings of 2 switches (SJA1110EVM)

The idea is to sent the traffic through two cascading switches, from talker to listener.

The four eth ports enabled Qbv are marked with red blocks in block diagram. 

As mentioned in previous post it is intended to create a GCL with cycle time 1ms = 125us * 8 slots.

Thanks a lot for help in advance!

BR,

He Wei

[abdullahzafar]

Hi Hewei,

You can revert admincycle time back to the one that you were using earlier(I think 62500). I earlier got confused between the two examples. 

I had a look at your configuration and it seems OK to me. I furthermore assume that the entire network is in sync or else, the other example would not have worked aswell. 

Following are the things that you should try:

1) Set the trigger time of the last entry to "1". That means that index 7 in every QBV table needs to be set to "1", everyother thing must remain unchange(Please view the attachment). Please first correct the admincycletime to your previous value before trying anything out. 

2) Read the drop counters, this will indicate on why the packets are being dropped. 

BR,

Abdullah