BGP Flow Specification Extensions for Network Congestion Management

Explicit Congestion Notification (ECN) is an extension to the Internet Protocol (IP) and the Transmission Control Protocol (TCP), defined in RFC 3168. ECN allows for end-to-end notification of congestion control to avoid packet loss. ECN is an optional feature that may be used by two endpoints if the underlying network infrastructure supports it. Typically, a TCP/IP network indicates a channel congestion by dropping packets. In the case of successful ECN negotiation, an ECN-aware router can set a flag in the IP header instead of dropping the packet to indicate that congestion is about to occur. The receiver of the packet then responds to the sender by reducing its transmission rate. The ECN (Explicit Congestion Notification) field in an IP header (bits 6 and 7 of the Traffic Class/ToS byte) allows routers to signal network congestion by marking packets instead of dropping them. Defined in RFC 3168, it uses four values (00-11) to indicate ECN capability and congestion status, enabling end-to-end congestion control while reducing packet loss and retransmissions. ECN Field Codepoints (Bits 6-7) 00 (Non-ECT): Non-ECN-Capable Transport. The packet does not support ECN. 01 or 10 (ECT): ECN-Capable Transport (0 or 1). The endpoints support ECN, and the path is not currently congested. 11 (CE): Congestion Experienced. The router is experiencing congestion and marks the packet, prompting the receiver to notify the sender to slow down. When a router's buffer exceeds a certain threshold, it changes the ECN field from ECT (01/10) to CE (11) rather than dropping the packet. It occupies the last two bits of the Type of Service (ToS) field in IPv4 or the Traffic Class field in IPv6. Fast-CNP (Congestion Notification Packet) is a mechanism for Remote Direct Memory Access (RoCEv2) networks that allows switches to directly send congestion notifications to the source node, bypassing the traditional, slower method of waiting for destination server feedback. By reducing latency in congestion signaling, Fast-CNP prevents severe buffer congestion and improves network performance. This document describes how to use BGP FlowSpec to implement network congestion management.

BGP-FS: BGP Flow Specification CE: Congestion Experienced DPI: Deep Packet Inspection ECN: Explicit Congestion Notification ECT: ECN Capable Transport FlowSpec: Flow Specification FSv2: BGP Flow Specification Version 2 SR: Segment Routing SR-MPLS: SR over the MPLS data plane SRv6: SR over the IPv6 data plane SAFI: Subsequent Address Family Identifier SID: Segment Identifier SRH: Segment Routing Header TE: Traffic Engineering USD: Ultimate Segment Decapsulation

Encoding: <type (1 octet), [numeric_op, value]+> Defines a list of {numeric_op, value} pairs used to match the 2-bit ECN field (see also [RFC3168] and [RFC8200]). This component uses the Numeric Operator (numeric_op) described in Section 4.2.1.1 of [RFC8955]. The ECN component values MUST be encoded as single octet (numeric_op len=00). The two least significant bits contain the ECN value. All other bits SHOULD be treated as 0. An example of an ECN Flow Specification component encoding for: "all packets matching ECN {1 or 2}".

The ecn marking Extended Community instructs a system to modify the ECN bits in the IP header (Section 5 of [RFC3168]) of a transiting IP packet to the corresponding value encoded in the 6 least significant bits of the Extended Community value, as shown in Figure XXX. The Extended Community is encoded as follows:

Type (1 octet): 0x80 Sub-Type (1 octet): TBD2 ECN: new ECN value for the transiting IP packet reserved (46 bits): MUST be set to 0 on encoding and MUST be ignored during decoding

The traffic-action Extended Community consists of 6 octets of which only the 2 least significant bits of the 6th octet (from left to right) are defined by [RFC8955], as shown in .

The S and T bits are defined in [RFC8955], and this document defines the C bit as follows: C Fast-CNP (bit 45): Enables the Fast CNP function (only effective when set).

The Fast-CNP Action Extended Community instructs a system to send fast congestion notification packets to the source node.

Type (1 octet): 0x80 Sub-Type (1 octet): TBD3 fast-cnp (1 octet): Enables the Fast CNP function if set to 1 and disables the Fast CNP function if set to 0 reserved (5 octets): MUST be set to 0 on encoding and MUST be ignored during decoding

The Marking Threshold Action Extended Community instructs a system to set a threshold for the target traffic.

Type (1 octet): 0x80 Sub-Type (1 octet): TBD4 lower threshold (2 octets): Lower threshold of the forwarding queue. upper threshold (2 octets): Upper threshold of the forwarding queue. reserved (7 bits): MUST be set to 0 on encoding and MUST be ignored during decoding. E (1 bit): Indicates whether to mark ECN flag.

A traffic management device on the network analyzes traffic on the network, and finds that there is a risk of packet loss due to congestion in traffic sent from Sender to Receiver through R1, R2,..., and Rn. The management device sends a Flowspec route to the device R1. The Flowspec route carries characteristic information of target traffic by using a Flowspec NLRI, and carries an marking ECN action defined in this draft.

When R1 receives a FlowSpec route, it uses the NLRI in the FlowSpec route to match the target traffic and marks the target traffic with the ECN flag according to the instruction of the ECN-Marking action. The traffic with the ECN flag is forwarded to the next hop. In this way, other devices along the path and the receiver can detect network congestion in a timely manner based on the ECN flag and make corresponding adjustments.

When a traffic management device in the network detects that traffic forwarded from a Sender terminal to a Receiver terminal through the network path R1, R2,..., Rn may encounter a risk of congestion and packet loss, the management device sends a Flowspec route to a device on the network path, for example, R2. The NLRI of the Flowspec route carries the characteristic of the target traffic, and the Flowspec route carries the Fast-CNP action defined in this draft.

After receiving the FlowSpec route, R2 uses the NLRI field in the FlowSpec route to match the target traffic. In addition, R2 identifies that the FlowSpec route carries the Fast-CNP action. According to the instruction of this action, R2 obtains the address of the sender terminal from the target traffic, constructs a Fast-CNP packet to be sent from R2 to the sender terminal, and sends the packet to the sender terminal. After receiving the Fast-CNP packet, the sender terminal reduces the rate at which packets are sent to the receiver terminal.

When the traffic management device in the network detects that traffic forwarding from a Sender terminal to a Receiver terminal through the network path R1, R2,..., Rn encounters congestion, the management device sends a Flowspec route to a device (for example, R1) on the network path. The NLRI of the Flowspec route carries the characteristic of the target traffic, where an ECN matching element is included, and the Flowspec route carries various possible traffic optimization actions. The traffic optimization action may include redirecting to a next hop, setting an optimized DCSP value, or indicating a forwarding through a path with a higher priority etc,.

After receiving the FlowSpec route, R1 matches the target traffic based on the NLRI field in the FlowSpec route. The ECN value in the FlowSpec route must be the same as that in the target traffic. After the target traffic is matched, R1 performs the traffic optimization action specified in the FlowSpec route on the target traffic. The traffic optimization action may include redirecting to a next hop, setting an optimized DCSP value, or indicating a forwarding through a path with a higher priority etc,.

When a management device on the network plans to optimize and manage target traffic on a device, the management device delivers a FlowSpec route to the network device (for example, R1 in the following figure). The FlowSpec route encapsulates the characteristics of the target traffic in the NLRI and uses the Marking Threshold Action community attribute defined in this document to set a threshold for the target traffic. In addition, the management device sets the E bit to instruct the network device to mark the target traffic with the ECN flag when the rate of the target traffic is within the threshold range.

When receiving the FlowSpec route, the network device R1 matches the target traffic based on the NLRI field in the FlowSpec route, sets a threshold for the target traffic according to the instruction of the marking-threshold action defined in this document, and enables the function of marking the target traffic with the ECN flag. The traffic with the ECN flag is forwarded to the next hop. In this way, other devices along the path and the receiver can detect network congestion in a timely manner based on the ECN flag and make corresponding adjustments.

TBD

The security considerations of BGP [RFC4271] and BGP FlowSpec [RFC8955] [RFC8956] apply to this document.

The following people made significant contributions to this document: TBD

The authors would like to acknowledge the review and inputs from XXX (TBD).