| Internet-Draft | WTR on Fast DF Recovery | July 2023 |
| Chen & Wang | Expires 11 January 2024 | [Page] |
The non-DF of a Single-Active Ethernet Segment (SA-ES) can filter both unicast traffic and BUM traffic. In such case, when a specific Single-Active ES performs fast DF recovery, the corresponding revertive FRR switching should be performed on the ingress PEs that are not adjacent to this ES. This revertive FRR switching needs to be performed immediately after the A-D per EVI route with the "P=1" is received from the new DF node. In other words, the revertive FRR switching cannot perform the WTR procedures, otherwise unicast traffic will be dropped on the new NDF node during the WTR.¶
In this draft, the SCT-EC is extended to the A-D per EVI routes, so that the ingress PEs can perform the revertive FRR switching based on the time specified by the SCT-EC for the non-adjacent ESes that support the fast DF recovery.¶
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.¶
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at https://datatracker.ietf.org/drafts/current/.¶
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."¶
This Internet-Draft will expire on 11 January 2024.¶
Copyright (c) 2023 IETF Trust and the persons identified as the document authors. All rights reserved.¶
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Revised BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Revised BSD License.¶
The non-DF of a Single-Active Ethernet Segment (SA-ES) can filter both unicast traffic and BUM traffic. In such case, when a specific Single-Active ES performs fast DF recovery, the corresponding revertive FRR switching should be performed on the ingress PEs that are not adjacent to this ES. This revertive FRR switching needs to be performed immediately after the A-D per EVI route with the "P=1" is received from the new DF node. In other words, the revertive FRR switching cannot perform the WTR procedures, otherwise unicast traffic will be dropped on the old DF (new NDF) node during the WTR.¶
However, on the other hand, lack of the WTR procedures also causes the unicast flows to be discarded on the newly activated path of the FRR until all corresponding forwarding entries on the newly activated path are completely installed. As a PE node that is not adjacent to a specified group of ESes, the ingress PE cannot know which ESes in that group are enabled with SCT-based fast DF recovery. If the ingress PEs stop executing the WTR procedures of the FRR of all non-adjacent ESes due to this reason, the traffic loss will be even worse.That is, the ESes that do not enable the SCT-based fast DF recovery at all will bear traffic loss due to the lack of WTR procedures.¶
In this draft, the DF Election Capabilities EC is extended to the A-D per ES routes, so that the ingress PEs can distinguish the non-adjacent ESes that support the fast DF recovery from the non-adjacent ESes that do not support the fast DF recovery. In this draft, the SCT-EC is extended to the A-D per EVI routes, so that the ingress PEs can perform the revertive FRR switching based on the time specified by the SCT-EC for the non-adjacent ESes that support the fast DF recovery.¶
The behavior of a non-DF AC is called NDF-behavior in this draft. The NDF-behavior of a SA-ES is different from the NDF-behavior of a All-Active ES.¶
The SA-ES topology is illustrated in the following figure. In that figure, the topology of Ethernet Segment ES21 is a MHN.¶
+---------+
+-------------+ | |
| | | |
CE1------+ (AC1) PE1 |----| | +-------------+
+ ^ | | | MPLS/ | | |---CE3
| | +-------------+ | VxLAN/ | | PE3 |
| ES21 | Cloud | | |
| | +-------------+ | |---| |
+ v | | | | +-------------+
CE2------+ (AC2) PE2 |----| |
| | | |
+-------------+ | |
+---------+
The DF filtering rules for SA-ES is different than the rules for AA-ES. The details are described in the following:¶
DF filtering for MHN is applied for traffic both ingress and egress on the access-facing Ethernet interfaces; whereas, DF filtering for AA-ES is applied only to traffic that egress the access-facing interfaces.¶
DF filtering for MHN impacts both unicast as well as flooded multi-destination traffic; whereas, DF filtering for AA-ES only applies to flooded multi-destination traffic..¶
Most of the terminology used in this documents comes from [I-D.ietf-bess-rfc7432bis] and [I-D.ietf-bess-evpn-fast-df-recovery] except for the following:¶
When a A-D per EVI route R1 is advertised to PE3 by PE1/PE2, PE3 will receive it after X ms. Now we can assume that X will be a random number from 1 ms to N ms in the network of Figure 1. In this example, we can assume that N will be as much as 60000 (ms) and the Ethernet Segment ES21 is a SA-ES.¶
In this example, the requirement is that, the packet-loss should be controlled at a less-than-10-ms level.¶
Based on the Service Carving Time (SCT) approach for Ethernet A-D per EVI Routes:¶
Initial state: The ES interface where AC2 is located is in steady state, and the ES interface where AC1 is located is is recovering. PE3 sends unicast flows to PE2 where AC2 is located.¶
The ES interface where the AC1 is located recovers at absolute time t=99.¶
PE1 advertises the RT-4 route (sent at the time t=100) for the ES interface where AC1 is located, carrying the target SCT value t=199. PE1 advertises the A-D per EVI route (sent at the time t=100) for AC1, carrying the target SCT value t=199 and "B=1" to the peer PE3 and PE2.¶
PE3 receives the A-D per EVI route on absolute time t=155.¶
PE3 executes FRR-revertion at absolute time t=199, while PE1 and PE2 execute service carving at absolute time t=199.¶
Using SCT approach, the negative effect of the lack of WTR procedures is mitigated. Furthermore, the BGP Ethernet Auto-discovery route (RT-1) transmission delay (from PE1 to PE3) becomes a non-issue. The use of SCT approach also remedies the problem associated with inconsistent pace between fast DF recovery (on PE1 and PE2) and FRR-revertion (on PE3): Revertive FRR switching on PE3 is now performed at the same absolute time as fast DF recovery on PE1 and PE2.¶
In case of fast DF recovery of a SA-ES, the Service Carving Timestamp Extended Community of [I-D.ietf-bess-evpn-fast-df-recovery] may be carried along with a A-D per EVI route.¶
1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Type = 0x06 | Sub-Type(0x0F)| Timestamp Seconds ~
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
~ Timestamp Seconds | Timestamp Fractional Seconds |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
The Time Synchronization bit (T bit) of [I-D.ietf-bess-evpn-fast-df-recovery] may also be carried along with the A-D per ES route before that SCT-EC is advertised . That T bit is carried in the DF Election Capabilities Extended Community per [RFC8584].¶
1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Type = 0x06 | Sub-Type(0x06)| RSV | DF Alg | |A| |T| ~
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
~ Bitmap | Reserved = 0 |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
In the case where one or more PEs attached to the Ethernet Segment do not signal T=1 in RT-4, all PEs in the Ethernet Segment SHALL revert back to the RT-1 per ES routes without the DF Election Capabilities Extended Community.¶
In the case where one of the non-adjacent PEs of the ES has signaled T=1 in its RT-1 per EVI, all PEs in that ES MAY advertise SCT-EC in their RT-1 per EVI routes. Although other PEs which are non-adjacent to that ES can't recognize the T bit in RT-1 per EVI route, their traffic loss will not be worse than current.¶
When AC1 recovers and PE1 decides to takeover the DF-role, and PE2 decides to takeover the NDF-role, PE1 will update its A-D per EVI Advertisement, and carry the SCT-EC and "B=1" along with that A-D per EVI route. PE2 will not update its previous A-D per EVI route (which carry the "P=1") until the time indicated by the SCT-EC.¶
When the SCT-EC of a A-D per EVI route still indicate a time in the future, the SCT-EC will not be used to select the best path for the corresponding ESI and ET-ID until the time indicated by the SCT-EC.¶
After the time indicated by the SCT-EC, PE3 will select the A-D per EVI route with that SCT-EC as the best path for the corresponding ESI and Ethernet Tag ID, no matter wthether the "B=1" or "P=1" is signaled along with that A-D per EVI.¶
When PE3 has two A-D per EVI routes for the same ESI and ET-ID, one of which carries the "P=1" and the other carries a SCT-EC whose timestamp indicates a time in the future, the route with the SCT-EC will not be selected as the best route until the time indicated by the SCT-EC. After the time indicated by the SCT-EC, the route with the SCT-EC will be selected as the best route.¶
When PE3 has two A-D per EVI routes for the same ESI and ET-ID, both of which carry the "B=1" along with themselves but one of which carries a SCT-EC whose timestamp indicates a time in the future, the route with the SCT-EC will not be selected as the best route until the time indicated by the SCT-EC.¶
Y ms after AC1 actually works as DF-role, PE1 should update the A-D per EVI route for AC1. In the updated A-D per EVI route, no SCT-EC should be included, and the P flag should be set.¶
It is possible that some non-adjacent PEs of a given ES continue to use the existing RT-1 per EVI route procedures and do not rely on the new SCT BGP extended community. PEs running a baseline RT-1 per EVI route procedures will simply discard the SCT BGP extended community as unrecognized.¶
A PE on a given ES can indicate its willingness to support clock-synched WTR by signaling the SCT-EC along with the RT-1 per EVI route. A PE which is not adjacent to that ES may discard the SCT BGP extended community as unrecognized, thus when the SCT-time comes that PE will do nothing because its "B=1" will not be updated before SCT-time. But a little bit later than the SCT-time, that PE will receive a RT-1 per EVI route with "P=1", and that RT-1 per EVI route will trigger the revertive FRR switching immediately. At least, this will not be worse than when the SCT-EC is not advertised.¶
There is no IANA consideration needed.¶