Help us improve your experience.

Let us know what you think.

Do you have time for a two-minute survey?

 
 

Understanding Priority-Based Flow Control

Priority-based flow control (PFC), IEEE standard 802.1Qbb, is a link-level flow control mechanism. The flow control mechanism is similar to that used by IEEE 802.3x Ethernet PAUSE, but it operates on individual priorities. Instead of pausing all traffic on a link, PFC allows you to selectively pause traffic according to its class.

This topic describes:

Reliability of Packet Delivery in Standard Ethernet Networks and in Layer 2 Networks

Standard Ethernet does not guarantee that a packet injected into the network will arrive at its intended destination. Reliability is provided by upper-layer protocols. Generally, a network path consists of multiple hops between the source and destination. A problem arises when transmitters send packets faster than receivers can accept them. When receivers run out of available buffer space to hold incoming flows, they silently drop additional incoming packets. This problem is generally resolved by upper-layer protocols that detect the drops and request retransmission.

Applications that require reliability in Layer 2 must have flow control that includes feedback from a receiver to a sender regarding buffer availability. Using IEEE 802.3x Ethernet PAUSE control frames, a receiver can generate a MAC control frame and send a PAUSE request to a sender when a specified threshold of receiver buffer has been filled to prevent buffer overflow. Upon receiving a PAUSE request, the sender stops transmission of any new packets until the receiver notifies the sender that it has sufficient buffer space to accept them again. The disadvantage of using Ethernet PAUSE is that it operates on the entire link, which might be carrying multiple traffic flows. Some traffic flows do not need flow control in Layer 2, because they are carrying applications that rely on upper-layer protocols for reliability. PFC enables you to configure Layer 2 flow control selectively for the traffic that requires it, such as Fibre Channel over Ethernet (FCoE) traffic, without impacting other traffic on the link. You can also enable PFC for other traffic types, such as iSCSI.

Calculations for Buffer Requirements When Using PFC PAUSE

The receive buffer must be large enough to accommodate all data that is received while the system is responding to a PFC PAUSE frame.

When you calculate buffer requirements, consider the following factors:

  • Processing and queuing delay of the PFC PAUSE—In general, the time to detect the lack of sufficient buffer space and to transmit the PFC PAUSE is negligible. However, delays can occur if the switch detects a reduction in buffer space just as the transmitter is beginning to transmit a maximum length frame.

  • Propagation delay across the media—The delay amount depends on the length and speed of the physical link.

  • Response time to the PFC PAUSE frame

  • Propagation delay across the media on the return path

Note:

We recommend that you configure at least 20 percent of the buffer size for the queue that is using PFC and that you do not specify the exact option.

Because it is mandatory to explicitly configure a certain percentage of buffer size for PFC, you must also explicity configure some buffer size for any other forwarding classes that you are planning to use (including the default forwarding classes and the user-defined forwarding classes). The percentage that you allocate depends on the usage of the respective classes.

How PFC and Congestion Notification Profiles Work With or Without DCBX

PFC can be applied to an interface regardless of whether the Data Center Bridging Capability Exchange protocol (DCBX) is enabled.

However, automatic control and advertisement of PFC requires DCBX:

  • When DCBX is enabled—DCBX detects the data center bridging (DCB) neighbor’s PFC configuration, uses autonegotiation to advertise local and peer PFC configuration, and then enables or disables PFC depending on whether the configurations are compatible or not. When PFC is enabled, it uses the congestion notification profile, which you have configured and applied to the interface.

  • When DCBX is not enabled—Class of service (CoS) triggers PFC when the incoming frame has a User Priority (UP) field that matches the three-bit pattern specified for the congestion notification profile.

To manually control the use of PFC on the interface regardless of the configuration of the peer data center devices, you can explicitly change the configuration of DCBX on the interface to disable PFC autonegotiation. See Disabling DCBX to Disable PFC Autonegotiation on EX Series Switches (CLI Procedure). When PFC autonegotiation is disabled, PFC is triggered by the congestion notification profile for PFC regardless of the configuration of the DCB peer.

Note:

PFC functions effectively only when the peer devices connected to the local interface are also using PFC and are configured compatibly with the local interface. PFC must be symmetrical—if PFC is not configured to use the same traffic class (code point) on both the local and the peer interface, it does not have any impact on the traffic.

Table 1 shows the one-to-one mapping between the UP field of an IEEE 802.1Q tagged frame, the traffic class, and the egress queue. In addition to setting a PFC congestion notification profile on an ingress port, you must set a forwarding class to match the priority specified in the PFC congestion notification profile and to forward the frame to the appropriate queue.

Juniper Networks EX Series Ethernet Switches support up to six traffic classes and allow you to associate those classes with six different congestion notification profiles. (The switches support up to 16 forwarding classes.)

Table 1: Input for PFC Congestion Notification Profile and Mapping to Traffic Class and Egress Queue

UP Field of IEEE-802.1Q Tagged Frame

Traffic Class

Egress Queue

000

TC 0

queue 0

001

TC 1

queue 1

010

TC 2

queue 2

011

TC 3

queue 3

100

TC4

queue 4

101

TC 5

queue 5