| RFC 8704 | BGP-Prefix SID in large-scale DCs | July 2019 |
| Filsfils, et al. | Informational | [Page] |
This document describes the motivation and benefits for applying segment routing in BGP-based large-scale data-centers. It describes the design to deploy segment routing in those data-centers, for both the MPLS and IPv6 dataplanes.¶
This document is not an Internet Standards Track specification; it is published for informational purposes.¶
This document is a product of the Internet Engineering Task Force (IETF). It represents the consensus of the IETF community. It has received public review and has been approved for publication by the Internet Engineering Steering Group (IESG). Not all documents approved by the IESG are candidates for any level of Internet Standard; see Section 2 of RFC 7841.¶
Information about the current status of this document, any errata, and how to provide feedback on it may be obtained at https://www.rfc-editor.org/info/rfc8704.¶
Copyright (c) 2019 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 Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.¶
Segment Routing (SR), as described in [I-D.ietf-spring-segment-routing] leverages the source routing paradigm. A node steers a packet through an ordered list of instructions, called segments. A segment can represent any instruction, topological or service-based. A segment can have a local semantic to an SR node or global within an SR domain. SR allows to enforce a flow through any topological path while maintaining per-flow state only at the ingress node to the SR domain. Segment Routing can be applied to the MPLS and IPv6 data-planes.¶
The use-cases described in this document should be considered in the context of the BGP-based large-scale data-center (DC) design described in [RFC7938]. This document extends it by applying SR both with IPv6 and MPLS dataplane.¶
This section provides a brief summary of the informational document [RFC7938] that outlines a practical network design suitable for data-centers of various scales:¶
The following is a schematic of a five-stage Clos topology, with four devices in the "Tier-1" stage. Notice that number of paths between Node1 and Node12 equals to four: the paths have to cross all of Tier-1 devices. At the same time, the number of paths between Node1 and Node2 equals two, and the paths only cross Tier-2 devices. Other topologies are possible, but for simplicity only the topologies that have a single path from Tier-1 to Tier-3 are considered below. The rest could be treated similarly, with a few modifications to the logic.¶
Tier-1
+-----+
|NODE |
+->| 5 |--+
| +-----+ |
Tier-2 | | Tier-2
+-----+ | +-----+ | +-----+
+------------>|NODE |--+->|NODE |--+--|NODE |-------------+
| +-----| 3 |--+ | 6 | +--| 9 |-----+ |
| | +-----+ +-----+ +-----+ | |
| | | |
| | +-----+ +-----+ +-----+ | |
| +-----+---->|NODE |--+ |NODE | +--|NODE |-----+-----+ |
| | | +---| 4 |--+->| 7 |--+--| 10 |---+ | | |
| | | | +-----+ | +-----+ | +-----+ | | | |
| | | | | | | | | |
+-----+ +-----+ | +-----+ | +-----+ +-----+
|NODE | |NODE | Tier-3 +->|NODE |--+ Tier-3 |NODE | |NODE |
| 1 | | 2 | | 8 | | 11 | | 12 |
+-----+ +-----+ +-----+ +-----+ +-----+
| | | | | | | |
A O B O <- Servers -> Z O O O
In the reference topology illustrated in Figure 1, It is assumed:¶
Each node is its own AS (Node X has AS X). 4-byte AS numbers are recommended ([RFC6793]).¶
Each node originates the IPv4 address of its loopback interface into BGP and announces it to its neighbors.¶
In this document, the Tier-1, Tier-2 and Tier-3 nodes are referred to respectively as Spine, Leaf and ToR (top of rack) nodes. When a ToR node acts as a gateway to the "outside world", it is referred to as a border node.¶
The data-center network design summarized above provides means for moving traffic between hosts with reasonable efficiency. There are few open performance and reliability problems that arise in such design:¶
First, it will be explained how to apply SR in the DC, for MPLS and IPv6 data-planes.¶
A BGP Prefix Segment is a segment associated with a BGP prefix. A BGP Prefix Segment is a network-wide instruction to forward the packet along the ECMP-aware best path to the related prefix.¶
The BGP Prefix Segment is defined as the BGP-Prefix-SID Attribute in [I-D.ietf-idr-bgp-prefix-sid] which contains an index. Throughout this document the BGP Prefix Segment Attribute is referred as the BGP-Prefix-SID and the encoded index as the label-index.¶
In this document, the network design decision has been made to assume that all the nodes are allocated the same SRGB (Segment Routing Global Block), e.g. [16000, 23999]. This provides operational simplification as explained in Section 8, but this is not a requirement.¶
For illustration purpose, when considering an MPLS data-plane, it is assumed that the label-index allocated to prefix 192.0.2.x/32 is X. As a result, a local label (16000+x) is allocated for prefix 192.0.2.x/32 by each node throughout the DC fabric.¶
When IPv6 data-plane is considered, it is assumed that Node X is allocated IPv6 address (segment) 2001:DB8::X.¶
Referring to Figure 1 and [RFC7938], the following design modifications are introduced:¶
Figure 2 zooms into a path from server A to server Z within the topology of Figure 1.¶
+-----+ +-----+ +-----+
+---------->|NODE | |NODE | |NODE |
| | 4 |--+->| 7 |--+--| 10 |---+
| +-----+ +-----+ +-----+ |
| |
+-----+ +-----+
|NODE | |NODE |
| 1 | | 11 |
+-----+ +-----+
| |
A <- Servers -> Z
Referring to Figure 1 and Figure 2 and assuming the IP address with the AS and label-index allocation previously described, the following sections detail the control plane operation and the data plane states for the prefix 192.0.2.11/32 (loopback of Node11)¶
Node11 originates 192.0.2.11/32 in BGP and allocates to it a BGP-Prefix-SID with label-index: index11 [I-D.ietf-idr-bgp-prefix-sid].¶
Node11 sends the following eBGP8277 update to Node10:¶
Node10 receives the above update. As it is SR capable, Node10 is able to interpret the BGP-Prefix-SID and hence understands that it should allocate the label from its own SRGB block, offset by the Label-Index received in the BGP-Prefix-SID (16000+11 hence 16011) to the NLRI instead of allocating a non-deterministic label out of a dynamically allocated portion of the local label space. The implicit-null label in the NLRI tells Node10 that it is the penultimate hop and must pop the top label on the stack before forwarding traffic for this prefix to Node11.¶
Then, Node10 sends the following eBGP8277 update to Node7:¶
Node7 receives the above update. As it is SR capable, Node7 is able to interpret the BGP-Prefix-SID and hence allocates the local (incoming) label 16011 (16000 + 11) to the NLRI (instead of allocating a "dynamic" local label from its label manager). Node7 uses the label in the received eBGP8277 NLRI as the outgoing label (the index is only used to derive the local/incoming label).¶
Node7 sends the following eBGP8277 update to Node4:¶
Node4 receives the above update. As it is SR capable, Node4 is able to interpret the BGP-Prefix-SID and hence allocates the local (incoming) label 16011 to the NLRI (instead of allocating a "dynamic" local label from its label manager). Node4 uses the label in the received eBGP8277 NLRI as outgoing label (the index is only used to derive the local/incoming label).¶
Node4 sends the following eBGP8277 update to Node1:¶
Node1 receives the above update. As it is SR capable, Node1 is able to interpret the BGP-Prefix-SID and hence allocates the local (incoming) label 16011 to the NLRI (instead of allocating a "dynamic" local label from its label manager). Node1 uses the label in the received eBGP8277 NLRI as outgoing label (the index is only used to derive the local/incoming label).¶
Referring to Figure 1, and assuming all nodes apply the same advertisement rules described above and all nodes have the same SRGB (16000-23999), here are the IP/MPLS forwarding tables for prefix 192.0.2.11/32 at Node1, Node4, Node7 and Node10.¶
| Incoming label or IP destination | Outgoing label | Outgoing Interface |
|---|---|---|
| 16011 | 16011 | ECMP{3, 4} |
| 192.0.2.11/32 | 16011 | ECMP{3, 4} |
| Incoming label or IP destination | Outgoing label | Outgoing Interface |
|---|---|---|
| 16011 | 16011 | ECMP{7, 8} |
| 192.0.2.11/32 | 16011 | ECMP{7, 8} |
| Incoming label or IP destination | Outgoing label | Outgoing Interface |
|---|---|---|
| 16011 | 16011 | 10 |
| 192.0.2.11/32 | 16011 | 10 |
| Incoming label or IP destination | Outgoing label | Outgoing Interface |
|---|---|---|
| 16011 | POP | 11 |
| 192.0.2.11/32 | N/A | 11 |
A network design choice could consist of switching all the traffic through Tier-1 and Tier-2 as MPLS traffic. In this case, one could filter away the IP entries at Node4, Node7 and Node10. This might be beneficial in order to optimize the forwarding table size.¶
A network design choice could consist in allowing the hosts to send MPLS-encapsulated traffic based on the Egress Peer Engineering (EPE) use-case as defined in [I-D.ietf-spring-segment-routing-central-epe]. For example, applications at HostA would send their Z-destined traffic to Node1 with an MPLS label stack where the top label is 16011 and the next label is an EPE peer segment ([I-D.ietf-spring-segment-routing-central-epe]) at Node11 directing the traffic to Z.¶
When the previous design is deployed, the operator enjoys global BGP-Prefix-SID and label allocation throughout the DC fabric.¶
A few examples follow:¶
The design previously described can be deployed incrementally. Let us assume that Node7 does not support the BGP-Prefix-SID and let us show how the fabric connectivity is preserved.¶
From a signaling viewpoint, nothing would change: even though Node7 does not support the BGP-Prefix-SID, it does propagate the attribute unmodified to its neighbors.¶
From a label allocation viewpoint, the only difference is that Node7 would allocate a dynamic (random) label to the prefix 192.0.2.11/32 (e.g. 123456) instead of the "hinted" label as instructed by the BGP-Prefix-SID. The neighbors of Node7 adapt automatically as they always use the label in the BGP8277 NLRI as outgoing label.¶
Node4 does understand the BGP-Prefix-SID and hence allocates the indexed label in the SRGB (16011) for 192.0.2.11/32.¶
As a result, all the data-plane entries across the network would be unchanged except the entries at Node7 and its neighbor Node4 as shown in the figures below.¶
The key point is that the end-to-end Label Switched Path (LSP) is preserved because the outgoing label is always derived from the received label within the BGP8277 NLRI. The index in the BGP-Prefix-SID is only used as a hint on how to allocate the local label (the incoming label) but never for the outgoing label.¶
| Incoming label or IP destination | Outgoing label | Outgoing interface |
|---|---|---|
| 12345 | 16011 | 10 |
| Incoming label or IP destination | Outgoing label | Outgoing interface |
|---|---|---|
| 16011 | 12345 | 7 |
The BGP-Prefix-SID can thus be deployed incrementally one node at a time.¶
When deployed together with a homogeneous SRGB (same SRGB across the fabric), the operator incrementally enjoys the global prefix segment benefits as the deployment progresses through the fabric.¶
The same exact design as eBGP8277 is used with the following modifications:¶
Cluster-1
+-----------+
| Tier-1 |
| +-----+ |
| |NODE | |
| | 5 | |
Cluster-2 | +-----+ | Cluster-3
+---------+ | | +---------+
| Tier-2 | | | | Tier-2 |
| +-----+ | | +-----+ | | +-----+ |
| |NODE | | | |NODE | | | |NODE | |
| | 3 | | | | 6 | | | | 9 | |
| +-----+ | | +-----+ | | +-----+ |
| | | | | |
| | | | | |
| +-----+ | | +-----+ | | +-----+ |
| |NODE | | | |NODE | | | |NODE | |
| | 4 | | | | 7 | | | | 10 | |
| +-----+ | | +-----+ | | +-----+ |
+---------+ | | +---------+
| |
| +-----+ |
| |NODE | |
Tier-3 | | 8 | | Tier-3
+-----+ +-----+ | +-----+ | +-----+ +-----+
|NODE | |NODE | +-----------+ |NODE | |NODE |
| 1 | | 2 | | 11 | | 12 |
+-----+ +-----+ +-----+ +-----+
For simple and efficient route propagation filtering and as illustrated in Figure 3:¶
The design described in [RFC7938] is reused with one single modification. It is highlighted using the example of the reachability to Node11 via spine node Node5.¶
Node5 originates 2001:DB8::5/128 with the attached BGP-Prefix-SID for IPv6 packets destined to segment 2001:DB8::5 ([I-D.ietf-idr-bgp-prefix-sid]).¶
Node11 originates 2001:DB8::11/128 with the attached BGP-Prefix-SID advertising the support of the SRH for IPv6 packets destined to segment 2001:DB8::11.¶
The control-plane and data-plane processing of all the other nodes in the fabric is unchanged. Specifically, the routes to 2001:DB8::5 and 2001:DB8::11 are installed in the FIB along the eBGP best-path to Node5 (spine node) and Node11 (ToR node) respectively.¶
An application on HostA which needs to send traffic to HostZ via only Node5 (spine node) can do so by sending IPv6 packets with a Segment Routing header (SRH, [I-D.ietf-6man-segment-routing-header]). The destination address and active segment is set to 2001:DB8::5. The next and last segment is set to 2001:DB8::11.¶
The application must only use IPv6 addresses that have been advertised as capable for SRv6 segment processing (e.g. for which the BGP prefix segment capability has been advertised). How applications learn this (e.g.: centralized controller and orchestration) is outside the scope of this document.¶
There are two general methods for communicating path information to the end-hosts: "proactive" and "reactive", aka "push" and "pull" models. There are multiple ways to implement either of these methods. Here, it is noted that one way could be using a centralized controller: the controller either tells the hosts of the prefix-to-path mappings beforehand and updates them as needed (network event driven push), or responds to the hosts making request for a path to specific destination (host event driven pull). It is also possible to use a hybrid model, i.e., pushing some state from the controller in response to particular network events, while the host pulls other state on demand.¶
It is also noted, that when disseminating network-related data to the end-hosts a trade-off is made to balance the amount of information Vs. the level of visibility in the network state. This applies both to push and pull models. In the extreme case, the host would request path information on every flow, and keep no local state at all. On the other end of the spectrum, information for every prefix in the network along with available paths could be pushed and continuously updated on all hosts.¶
As required by [RFC7938], no new signaling protocol is introduced. The BGP-Prefix-SID is a lightweight extension to BGP Labeled Unicast [RFC8277]. It applies either to eBGP or iBGP based designs.¶
Specifically, LDP and RSVP-TE are not used. These protocols would drastically impact the operational complexity of the Data Center and would not scale. This is in line with the requirements expressed in [RFC7938].¶
Provided the same SRGB is configured on all nodes, all nodes use the same MPLS label for a given IP prefix. This is simpler from an operation standpoint, as discussed in Section 8¶
The designer may decide to switch all the traffic at Tier-1 and Tier-2's based on MPLS, hence drastically decreasing the IP table size at these nodes.¶
This is easily accomplished by encapsulating the traffic either directly at the host or the source ToR node by pushing the BGP-Prefix-SID of the destination ToR for intra-DC traffic, or the BGP-Prefix-SID for the the border node for inter-DC or DC-to-outside-world traffic.¶
It is straightforward to combine the design illustrated in this document with the Egress Peer Engineering (EPE) use-case described in [I-D.ietf-spring-segment-routing-central-epe].¶
In such case, the operator is able to engineer its outbound traffic on a per host-flow basis, without incurring any additional state at intermediate points in the DC fabric.¶
For example, the controller only needs to inject a per-flow state on the HostA to force it to send its traffic destined to a specific Internet destination D via a selected border node (say Node12 in Figure 1 instead of another border node, Node11) and a specific egress peer of Node12 (say peer AS 9999 of local PeerNode segment 9999 at Node12 instead of any other peer which provides a path to the destination D). Any packet matching this state at host A would be encapsulated with SR segment list (label stack) {16012, 9999}. 16012 would steer the flow through the DC fabric, leveraging any ECMP, along the best path to border node Node12. Once the flow gets to border node Node12, the active segment is 9999 (because of PHP on the upstream neighbor of Node12). This EPE PeerNode segment forces border node Node12 to forward the packet to peer AS 9999, without any IP lookup at the border node. There is no per-flow state for this engineered flow in the DC fabric. A benefit of segment routing is the per-flow state is only required at the source.¶
As well as allowing full traffic engineering control such a design also offers FIB table minimization benefits as the Internet-scale FIB at border node Node12 is not required if all FIB lookups are avoided there by using EPE.¶
The design presented in this document preserves the availability and load-balancing properties of the base design presented in [I-D.ietf-spring-segment-routing].¶
For example, one could assign an anycast loopback 192.0.2.20/32 and associate segment index 20 to it on the border Node11 and Node12 (in addition to their node-specific loopbacks). Doing so, the EPE controller could express a default "go-to-the-Internet via any border node" policy as segment list {16020}. Indeed, from any host in the DC fabric or from any ToR node, 16020 steers the packet towards the border Node11 or Node12 leveraging ECMP where available along the best paths to these nodes.¶
In the MPLS case, it is recommend to use same SRGBs at each node.¶
Different SRGBs in each node likely increase the complexity of the solution both from an operational viewpoint and from a controller viewpoint.¶
From an operation viewpoint, it is much simpler to have the same global label at every node for the same destination (the MPLS troubleshooting is then similar to the IPv6 troubleshooting where this global property is a given).¶
From a controller viewpoint, this allows us to construct simple policies applicable across the fabric.¶
Let us consider two applications A and B respectively connected to Node1 and Node2 (ToR nodes). A has two flows FA1 and FA2 destined to Z. B has two flows FB1 and FB2 destined to Z. The controller wants FA1 and FB1 to be load-shared across the fabric while FA2 and FB2 must be respectively steered via Node5 and Node8.¶
Assuming a consistent unique SRGB across the fabric as described in the document, the controller can simply do it by instructing A and B to use {16011} respectively for FA1 and FB1 and by instructing A and B to use {16005 16011} and {16008 16011} respectively for FA2 and FB2.¶
Let us assume a design where the SRGB is different at every node and where the SRGB of each node is advertised using the Originator SRGB TLV of the BGP-Prefix-SID as defined in [I-D.ietf-idr-bgp-prefix-sid]: SRGB of Node K starts at value K*1000 and the SRGB length is 1000 (e.g. Node1's SRGB is [1000, 1999], Node2's SRGB is [2000, 2999], ...).¶
In this case, not only the controller would need to collect and store all of these different SRGB's (e.g., through the Originator SRGB TLV of the BGP-Prefix-SID), furthermore it would need to adapt the policy for each host. Indeed, the controller would instruct A to use {1011} for FA1 while it would have to instruct B to use {2011} for FB1 (while with the same SRGB, both policies are the same {16011}).¶
Even worse, the controller would instruct A to use {1005, 5011} for FA1 while it would instruct B to use {2011, 8011} for FB1 (while with the same SRGB, the second segment is the same across both policies: 16011). When combining segments to create a policy, one need to carefully update the label of each segment. This is obviously more error-prone, more complex and more difficult to troubleshoot.¶
This document does not make any IANA request.¶
The design and deployment guidelines described in this document are based on the network design described in [RFC7938].¶
The deployment model assumed in this document is based on a single domain where the interconnected DCs are part of the same administrative domain (which, of course, is split into different autonomous systems). The operator has full control of the whole domain and the usual operational and management mechanisms and procedures are used in order to prevent any information related to internal prefixes and topology to be leaked outside the domain.¶
As recommended in [I-D.ietf-spring-segment-routing], the same SRGB should be allocated in all nodes in order to facilitate the design, deployment and operations of the domain.¶
When EPE ([I-D.ietf-spring-segment-routing-central-epe]) is used (as explained in Section 7.3, the same operational model is assumed. EPE information is originated and propagated throughout the domain towards an internal server and unless explicitly configured by the operator, no EPE information is leaked outside the domain boundaries.¶
This document proposes to apply Segment Routing to a well known scalability requirement expressed in [RFC7938] using the BGP-Prefix-SID as defined in [I-D.ietf-idr-bgp-prefix-sid].¶
It has to be noted, as described in Section 10 that the design illustrated in [RFC7938] and in this document, refer to a deployment model where all nodes are under the same administration. In this context, it is assumed that the operator doesn't want to leak outside of the domain any information related to internal prefixes and topology. The internal information includes prefix-sid and EPE information. In order to prevent such leaking, the standard BGP mechanisms (filters) are applied on the boundary of the domain.¶
Therefore, the solution proposed in this document does not introduce any additional security concerns from what expressed in [RFC7938] and [I-D.ietf-idr-bgp-prefix-sid]. It is assumed that the security and confidentiality of the prefix and topology information is preserved by outbound filters at each peering point of the domain as described in Section 10.¶
The authors would like to thank Benjamin Black, Arjun Sreekantiah, Keyur Patel, Acee Lindem and Anoop Ghanwani for their comments and review of this document.¶
Gaya Nagarajan Facebook US Email: gaya@fb.com Gaurav Dawra Cisco Systems US Email: gdawra.ietf@gmail.com Dmitry Afanasiev Yandex RU Email: fl0w@yandex-team.ru Tim Laberge Cisco US Email: tlaberge@cisco.com Edet Nkposong Salesforce.com Inc. US Email: enkposong@salesforce.com Mohan Nanduri Microsoft US Email: mnanduri@microsoft.com James Uttaro ATT US Email: ju1738@att.com Saikat Ray Unaffiliated US Email: raysaikat@gmail.com Jon Mitchell Unaffiliated US Email: jrmitche@puck.nether.net¶