Users Guide

How Priority-Based Flow Control is Implemented
Priority-based ow control provides a ow control mechanism based on the 802.1p priorities in converged Ethernet trac received on an
interface and is enabled by default. As an enhancement to the existing Ethernet pause mechanism, PFC stops trac transmission for
specied priorities (CoS values) without impacting other priority classes. Dierent trac types are assigned to dierent priority classes.
When trac congestion occurs, PFC sends a pause frame to a peer device with the CoS priority values of the trac that needs to be
stopped. DCBx provides the link-level exchange of PFC parameters between peer devices. PFC creates zero-loss links for SAN trac that
requires no-drop service, while at the same time retaining packet-drop congestion management for LAN trac.
PFC is implemented on an Aggregator as follows:
• If DCB is enabled, as soon as a dcb-map with PFC is applied on an interface, DCBx starts exchanging information with PFC-enabled
peers. The IEEE802.1Qbb, CEE and CIN versions of PFC TLV are supported. DCBx also validates PFC congurations received in TLVs
from peer devices.
• To achieve complete lossless handling of trac, enable PFC operation on ingress port trac and on all DCB egress port trac.
• All 802.1p priorities are enabled for PFC. Queues to which PFC priority trac is mapped are lossless by default. Trac may be
interrupted due to an interface ap (going down and coming up).
• For PFC to be applied on an Aggregator port, the auto-congured priority trac must be supported by a PFC peer (as detected by
DCBx).
• A dcb-map for PFC applied to an interface may become invalid if dot1p-queue mapping is recongured. This situation occurs when the
new dot1p-queue assignment exceeds the maximum number (2) of lossless queues supported globally on the switch. In this case, all
PFC congurations received from PFC-enabled peers are removed and re-synchronized with the peer devices.
• Dell Networking OS does not support MACsec Bypass Capability (MBC).
Conguring Lossless Queues
DCB also supports the manual conguration of lossless queues on an interface when PFC mode is turned o and priority classes are
disabled in a DCB map, apply the map on the interface.
Prerequisite: A DCB input policy with PFC conguration is applied to the interface with the following conditions:
• PFC mode is o (no pfc mode on).
• No PFC priority classes are congured (no pfc priority priority-range).
Example:
Port A —> Port B
Port C —> Port B
PFC no-drop queues are congured for queues 1, 2 on Port B. PFC capability is enabled on priorities 3, 4 on PORT A and C.
Port B acting as Egress
During the congestion, [trac pump on priorities 3 and 4 from PORT A and PORT C is at full line rate], PORT A and C send out the PFCs
to rate the trac limit. Egress drops are not observed on Port B since trac ow on priorities is mapped to loss less queues.
Port B acting as Ingress
If the trac congestion is on PORT B , Egress DROP is on PORT A or C, as the PFC is not enabled on PORT B.
Refer the following conguration for queue to dot1p mapping:
Dell(conf)#do show qos dot1p-queue-mapping
Dot1p Priority : 0 1 2 3 4 5 6 7 -> On ingress interfaces[Port A and C] we used the
PFC on priority level.
Queue : 0 0 0 1 2 3 3 3 -> On Egress interface[Port B] we used no-drop queues.
Data Center Bridging (DCB)
257