Some of the TGEN used for the testing along with SNAPPI have default queue size of 4.
With 8 queues, it is easy to map the PRIO flows with the hardware queues (usually one-one mapping).
However, with less than 8 queues, could result in different results. With TGEN, that we used, it was observed that flows with PRIO4 and above, did not have any hardware queues assigned to them. This resulted in improper behavior especially when PFC (for Prio4) was involved during the test.
Issue Description
Some of the TGEN used for the testing along with SNAPPI have default queue size of 4.
With 8 queues, it is easy to map the PRIO flows with the hardware queues (usually one-one mapping).
However, with less than 8 queues, could result in different results. With TGEN, that we used, it was observed that flows with PRIO4 and above, did not have any hardware queues assigned to them. This resulted in improper behavior especially when PFC (for Prio4) was involved during the test.
Current code: pfc.pfc_class_0 = 0 pfc.pfc_class_1 = 1 pfc.pfc_class_2 = 2 pfc.pfc_class_3 = 3 pfc.pfc_class_4 = 4 pfc.pfc_class_5 = 5 pfc.pfc_class_6 = 6 pfc.pfc_class_7 = 7
There should check in script to queue lengths and allow for dynamic queue vs prio assignments.
@kamalsahu0001
Results you see
We observed that PFCs for 4 were not getting counted on the TGEN and usually, it reacted to the PFCs received for PRIO3.
Results you expected to see
We have a workaround for now, to assign specific queues for Prio4 and above.
Is it platform specific
generic
Relevant log output
No response
Output of
show version
No response
Attach files (if any)
No response