Is anyone else tracking discards on the back-end switch ports?

We recently replaced our backend Infiniband switches with Dell EMC OS10 Ethernet switches (S5232F-ON)

On one of them, the ifDiscard counters are mostly at zero; on the other, they run at ~.3 -- 1.2% of the traffic (despite trivial levels of traffic)

I also notice that utilization is heavily skewed toward one switch (fortunately, the one reporting near-zero discards).

Anyone else see patterns like this?

These discards rates seem unusually high to me, compared to the other Data Center switches we manage ... even the front-end switches to this cluster report hardly any discards, despite far higher traffic levels.

--sk

Stuart Kendrick