Administrator Guide

Priority-Based Flow Control
In a data center network, priority-based ow control (PFC) manages large bursts of one trac type in multiprotocol links so that it
does not aect other trac types and no frames are lost due to congestion.
When PFC detects congestion on a queue for a specied priority, it sends a pause frame for the 802.1p priority trac to the
transmitting device. In this way, PFC ensures that large amounts of queued LAN trac do not cause storage trac to be dropped,
and that storage trac does not result in high latency for high-performance computing (HPC) trac between servers.
PFC enhances the existing 802.3x pause and 802.1p priority capabilities to enable ow control based on 802.1p priorities (classes of
service). Instead of stopping all trac on a link (as performed by the traditional Ethernet pause mechanism), PFC pauses trac on a
link according to the 802.1p priority set on a trac type. You can create lossless ows for storage and server trac while allowing for
loss in case of LAN trac congestion on the same physical interface.
The following illustration shows how PFC handles trac congestion by pausing the transmission of incoming trac with dot1p
priority 3.
Figure 1. Priority-Based Flow Control
In the system, PFC is implemented as follows:
PFC is supported on specied 802.1p priority trac (dot1p 0 to 7) and is congured per interface. However, only two lossless
queues are supported on an interface: one for Fibre Channel over Ethernet (FCoE) converged trac and one for Internet Small
Computer System Interface (iSCSI) storage trac. Congure the same lossless queues on all ports.
A dynamic threshold handles intermittent trac bursts and varies based on the number of PFC priorities contending for buers,
while a static threshold places an upper limit on the transmit time of a queue after receiving a message to pause a specied
priority. PFC trac is paused only after surpassing both static and dynamic thresholds for the priority specied for the port.
By default, PFC is enabled when you enabled DCB. When you enable DCB globally, you cannot simultaneously enable TX and RX
on the interface for ow control and link-level ow control is disabled.
Buer space is allocated and de-allocated only when you congure a PFC priority on the port.
PFC delay constraints place an upper limit on the transmit time of a queue after receiving a message to pause a specied priority.
By default, PFC is enabled on an interface with no dot1p priorities congured. You can congure the PFC priorities if the switch
negotiates with a remote peer using DCBX. During DCBX negotiation with a remote peer:
DCBx communicates with the remote peer by link layer discovery protocol (LLDP) type, length, value (TLV) to determine
current policies, such as PFC support and enhanced transmission selection (ETS) BW allocation.
If the negotiation succeeds and the port is in DCBX Willing mode to receive a peer conguration, PFC parameters from the
peer are used to congured PFC priorities on the port. If you enable the link-level ow control mechanism on the interface,
DCBX negotiation with a peer is not performed.
If the negotiation fails and PFC is enabled on the port, any user-congured PFC input policies are applied. If no PFC dcb-map
has been previously applied, the PFC default setting is used (no priorities congured). If you do not enable PFC on an
interface, you can enable the 802.3x link-level pause function. By default, the link-level pause is disabled, when you disable
DCBx and PFC. If no PFC dcb-map has been applied on the interface, the default PFC settings are used.
PFC supports buering to receive data that continues to arrive on an interface while the remote system reacts to the PFC
operation.
Data Center Bridging (DCB)
27