[c-nsp] PAUSE vs PFC for loss-less traffic?

[Mathias Sundman]

There is a lot of talk about how PFC makes Ethernet loss-less and suitable
for FCoE or ISCSI. My question is, for single-type-of-traffic interfaces,
for example ISCSI arrays connected to a Ethernet switch. Why doesn't
traditional PAUSE based flow control give the same level of loss-less
performance?

For example Dell claims in section 5.1 in their guide "Creating a DCB
Compliant EqualLogic iSCSI SAN with Mixed Traffic" that retransmissions
went from 0.315% to 0% when going from PAUSE to PFC. I just don't
understand why. Both technologies works the same way with pause for xx
time, and resume frames. Are the thresholds for when to send the pause
frames different by default?

Reason for asking is that Dell does not support running DCB between Nexus
switches and Equallogic arrays, and I'm trying to understand why I couldn't
just do traditional PAUSE based flow control on the interfaces between the
array and the switches, as those are ISCSI only interfaces and manually map
the traffic into the correct qos-group, and then do DCB with PFC on the
server interfaces to have flow control only on the ISCSI VLAN.

Any thoughts?