Reference Guide
Generation of PFC for a Priority for Untagged Packets
In order to generate PFC for a particular priority for untagged packets, and configuring PFC for that priority, you should find the
queue number associated with priority from TABLE 1 and Associate a DCB map to forward the matched DSCP packet to that
queue. PFC frames gets generated with PFC priority associated with the queue when the queue gets congested.
Operations on Untagged Packets
The below is example for enabling PFC for priority 2 for tagged packets. Priority (Packet Dot1p) 2 will be mapped to PG6 on
PRIO2PG setting. All other Priorities for which PFC is not enabled are mapped to default PG – PG7.
Classification rules on ingress (Ingress FP CAM region) matches incoming packet-dot1p and assigns an internal priority (to
select queue as per Table 1 and Table 2).
The internal Priority assigned for the packet by Ingress FP is used by the memory management unit (MMU) to assign the packet
to right queue by indexing the internal-priority to queue map table (TABLE 1) in hardware.
PRIO2COS setting for honoring the PFC protocol packets from the Peer switches is as per above Packet-Dot1p->queue table
(Table 2).
The packets that come in with packet-dot1p 2 alone will be assigned to PG6 on ingress.
The packets that come in with packet-dot1p 2 alone will use Q1 (as per dot1p to Queue classification – Table 2) on the egress
port.
● When Peer sends a PFC message for Priority 2, based on above PRIO2COS table (TABLE 2), Queue 1 is halted.
● Queue 1 starts buffering the packets with Dot1p 2. This causes PG6 buffer counter to increase on the ingress, since P-dot1p
2 is mapped to PG6.
● As the PG6 watermark threshold is reached, PFC will be generated for dot1p 2.
Performing PFC Using DSCP Bits Instead of 802.1p
Bits
Priority based Flow Control (PFC) is currently supported on Dell Networking OS for tagged packets based on the packet Dot1p.
In certain data center deployments, VLAN configuration is avoided on the servers and all packets from the servers are untagged.
These packets will carry IP header and can be differentiated based on the DSCP fields they carry on the server facing switch
ports. Requirement is to classify these untagged packets from the server based on their DSCP and provide PFC treatment.
Dell Networking OS Releases 9.3(0.0) and earlier provide CLI support to specify the priorities for which PFC is enabled on each
port. This feature is applicable only for the tagged packets based on the incoming packet Dot1p and Dot1p based queue
classification. This document will discuss the configurations required to support PFC for untagged packets based on incoming
packet DSCP.
For the tagged packets, Queue is selected based on the incoming Packet Dot1p. When PFC frames for a specific priority is
received from the peer switch, the queue corresponding to that Dot1p is halted from scheduling on that port, thus honoring the
PFC from the peer. If a queue is congested due to packets with a specific Dot1p and PFC is enabled for that Dot1p, switch will
transit out PFC frames for that Dot1p. The packet Dot1p to Queue mapping for classification on the ingress must be same as the
mapping of Dot1p to the Queue to be halted on the egress used for PFC honoring. Dell Networking OS ensures that these
mappings are identical. This section discusses the Dell Networking OS configurations needed for above PFC generation and
honoring mechanism to work for the untagged packets.
PRIORITY to PG mapping (PRIO2PG) is on the ingress for each port. By default, all priorities are mapped to PG7. A priority for
which PFC has to be generated is assigned to a PG other than PG7 (say PG6) and buffer watermark is set on PG6 so as to
generate PFC.
In ingress, the buffers are accounted at per PG basis and would indicate the number of the packets that has ingress this port
PG but still queued up in egress pipeline. However, there is no direct mapping between the PG and Queue.
Packet is assigned an internal priority on the ingress pipeline based on the queue to which it is destined. This Internal-priority to
Queue mapping has been modified and enhanced as follows for the device:
234
Data Center Bridging (DCB)