PFC Pause Frames on FEX Port & VPC Peer Link

[thecompnerd]

Is it normal to see PFC pause frames on a FEX fabric port?  The 2148T FEX or any FEX connected devices shouldn't even be PFC capable from my understanding.  So why would I see pause frames on the fabric port?

SWITCH# show interface eth1/5 priority-flow-control
============================================================
Port               Mode Oper(VL bmap)  RxPPP      TxPPP
============================================================

Ethernet1/5        Off  Off           1147675    0

Additionally, there are tons of RX and TX PFC pause frames on the VPC peer link.  In the peer link, average utilization is about 15% on each physical member.  Unless there are large spikes of data I'm not catching, I don't get why I'm seeing so many pause frames.  I know they're to be expected as we have DCB enabled since we have iSCSI traffic on the network.  However, we're seeing 100s of thousands of pause frames in about a 12 hour period which seems excessive.

The only reason why I'm even focusing on these pause frames is that we're seeing iSCSI disconnects in ESX hosts.  The server/storage vendor is concerned about the pause frames causing the issue.  I have no way of disproving that theory at this time.

[will]

hi, just now saw this after reviewing my own DCB problems. that FEX does appear to support PFC:

http://www.cisco.com/c/en/us/products/collateral/switches/nexus-2000-series-fabric-extenders/data_sheet_c78-507093.html

However, it looks like the pause frames are coming from the server/storage CNA. this would imply that this device has a default policy of sending pause frames for certain priorities and this policy is active. I think this could be the case, even if the switch side is not configured. I would expect the switch to just ignore this PFC messaging from the CNA, meaning from the switch perspective its doing nothing.

try this command to see if PFC is enabled on the interface on the nexus:

switch# show interface priority-flow-control

the big question is why is the device sending so many PFC's back to the switch. sounds the CNA is getting overwhelmed by the sending device / nexus switchport, which is trying to send it "too much data".

[joaquin.romero]

Will,

Saw this post and wanted to ask you what your root cause of the the CNA sending high RXPPP was? We are currently seeing the same situation ([quantity 2] B420 M3s sending a high number of Pause in both TX an RX).

Any insight into what you found when your host was exhibiting the behavior would be most appreciated here.

Thanks.