| SPRING Working Group | S. Litkowski |
| Internet-Draft | Orange |
| Intended status: Informational | M. Aissaoui |
| Expires: February 10, 2018 | Nokia |
| August 9, 2017 |
Implementing non protected paths using SPRING
draft-litkowski-spring-non-protected-paths-02
Segment Routing (SR) leverages the source routing paradigm. A node can steer a packet on a specific path by prepending the packet with an SR header. In the framework of traffic-engineering use cases, a customer may request its service provider to implement some non protected paths. This means that in case of a failure within the network, fast-reroute (or similar) techniques should not be activated for those paths. This document analyzes the different options to implement a non protected path with Segment Routing and in a future release will provide a recommandation on the best option.
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in [RFC2119].
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 http://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 February 10, 2018.
Copyright (c) 2017 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 (http://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.
In some cases, a customer may prefer to react on network failures using its own mechanism. In such cases, the customer usually has two disjoint paths, so a path can take over the traffic in case of failure of the other. The disjoint paths can be provided by a single provider or by multihoming to different providers as displayed in the figure below.
_________________________________________
/ \
/ \
| |
| |
| |
| ***********************> |
| +------+ 10 +------+ |
CE1 ****| PE 1 | ----- R1 ---- R2 ------- | PE 2 |**** CE2
| +------+ | | +------+ |
| | | |
| | | |
| +------+ | | +------+ |
CE3 ****| PE 3 | ----- R3 ---- R4 ------- | PE 4 |**** CE4
| +------+ ***********************> +------+ |
| |
\ /
\_________________________________________/
Figure 1 - Disjoint paths provided by a single provider
_________________________________________
/ \
/ \
| |
| |
| ***********************> |
| +------+ 10 +------+ |
CE1 ****| PE 1 | ----- R1 ---- R2 ------- | PE 2 |**** CE2
| +------+ +------+ |
| |
\ /
\_________________________________________/
_________________________________________
/ \
/ \
| |
| |
| +------+ +------+ |
CE3 ****| PE 1 | ----- R3 ---- R4 ------- | PE 2 |**** CE4
| +------+ ***********************> +------+ |
| |
\ /
\_________________________________________/
Figure 2 - Disjoint paths provided by using two providers
As the traffic protection is ensured by an end-to-end mechanism at the customer level, the customer requests the service provider to not protect the paths. This is particularly required to avoid both protection mechanisms (customer level and provider level) to be activated at the same time which may lead to unpredictable side effects. However the service provider is allowed to restore the end-to-end path automatically when the primary path is failing by computing and installing a new primary path at the head-end. How the end-to-end protection is handled is out of scope of this document and will be under the customer responsibility.
Another use case could be a service provider selling the traffic protection as a service option. So by default, the provided IP/MPLS path is not protected by any fast-reroute mechanism but the customer can subscribe to an option to activate fast-reroute for its traffic. In the figure 3, the Customer1 service between PE1 and PE2 is protected, in case of failure between R1 and R2, the LSP can use a bypass through R3-R4 nodes until the convergence occurs. The Customer2 did not subscribe to the traffic protection option. If R3-R4 fails, the traffic between CE3 and CE4 will be disrupted until the convergence occurs.
_________________________________________
/ \
/ \
| Protected LSP |
| ***********************> |
| +------+ 10 +------+ |
CE1 ****| PE 1 | ----- R1 ---- R2 ------- | PE 2 |**** CE2 Customer1
| +------+ |* *| +------+ |
| |* *| |
| |* *| |
| +------+ |********| +------+ |
CE3 ****| PE 3 | ----- R3 ---- R4 ------- | PE 4 |**** CE4 Customer2
| +------+ ***********************> +------+ |
| Non protected LSP |
\ /
\_________________________________________/
Figure 3 - Provider selling traffic protection as an option
A service provider may also propose a traffic protection service based on path protection rather than local repair on each transit node. In the figure 4, on PE1, two LSPs were created to ensure the customer traffic protection between PE1 and PE2. The primary LSP is used to carry the traffic in the nominal situation. The protection LSP is built as disjoint from the primary LSP and may be preestablished (from controlplane and/or dataplane point of view). When the primary LSP fails, PE1 is responsible to switch the traffic to the protection LSP. As the protection is provided by PE1, both primary and protection LSPs should be setup as non protected so transit nodes will not activate any local-repair mechanism for those LSPs.
_________________________________________
/ \
/ \
| Primary LSP |
| ***********************> |
| +------+ 10 +------+ |
CE1 ****| PE 1 | ----- R1 ---- R2 ------- | PE 2 |**** CE2 Customer1
| +------+ | | +------+ |
| *\ | | /> |
| *\ | | /* |
| *\ | | /* |
| *+- R3 ---- R4 ----+* |
| ******************* |
| Protection LSP |
\ /
\_________________________________________/
Figure 4 - Provider selling traffic protection as an option
A segment-routing path is expressed as a list of segment identifiers (SID) from different types (Node-SID, Adj-SID, Binding-SID ...). In order to ensure that the segment routing path is not protected, we need to ensure that it does not contain any segment representing a protected path. As an example, in the Figure 1, we consider a path from PE1 to PE2 expressed with the following segment list: {Adj_R1R3,Node_R2,Adj_R2PE2}. If we want to ensure that this path is not protected, we need to ensure that the segment represented by Adj_R1R3 represents a non protected segment, as well as the segments Node_R2 and Adj_R2PE2.
The segment routing path may be computed by a Path Computation Element (PCE). In order to fulfil the non protected path constraint, the PCE needs to be aware of the available SIDs in the network and their protection status.
Several techniques may be used to represent a non protected path with a segment identifier. We propose to analyze the different options.
When equal cost paths are available within the end-to-end path, implementations may reuse a fast-reroute like mechanism in the dataplane, so when one of the outgoing interface fails, the dataplane switches traffic immediately to the remaining outgoing interfaces in the ECMP set. This behavior is usually hardcoded and cannot be disabled. Based on this assumption, a non protected LSP SHOULD avoid ECMPs.
A node can advertise multiple adjacency segments for a particular link with different properties. The non-protected property is already defined as part of the protocol encodings ([I-D.ietf-isis-segment-routing-extensions], [I-D.ietf-ospf-segment-routing-extensions] and [I-D.gredler-idr-bgp-ls-segment-routing-extension]) through the B flag. However, from an implementation perspective, advertising a protected adjacency segment, a non protected adjacency segment or both for each link is optional.
It is important to note that even if an adjacency segment has the B flag set (protected), it remains up to a local policy of the advertising router to implement the protection or not.
If both protected and non protected Adj-SID are advertised, every node in the network (including PCEs) can be aware of the adjacency segments protection property. When a non protected path is requested, the path computation module can choose to encode the path with a list a non protected adjacency segments only.
One of the advantage of using only adjacency segments is the insurance that the traffic will never go transiently outside the path defined by the computation module responsible of the path. This solution is fully compliant with the requirements sets in Section 2.
One of the drawbacks of using only adjacency segments is the resulting label stack depth as each hop should require a segment in the stack: crossing 15 nodes, means stacking 15 labels to encode the SR tunnel. Having such a deep stack may be a problem for current hardwares and softwares for either pushing the stack (because the head end is limited in the number of labels it can push) or loadbalancing flows on transit nodes (as deep packet inspection or entropy label look up may be difficult with a deep label stack). Another drawback of advertising both protected and non protected adjacency segments is the additional controlplane and dataplane resource consumption used in the network. As the adjacency SIDs have a local significance, this resource consumption can be considered as negligeable from a data plane point of view. From a control plane point of view, this can also be considered as negligeable with the current CPU and memory usually available on routers.
Using a combination of node segments and adjacency segments is the usual way of creating a segment routing path. However the well known Node-SID (algorithm type Shortest Path) may be protected by a local-repair mechanism by any transit node or may use ECMPs which may be a problem when used for a non protected path. Protecting a particular Node-SID is a matter of a local policy configuration on every node. The following discusses a number of possible approaches.
As for adjacency segments, a new flag may be added in the Prefix-SID to encode the willingness of protection. Each node will then advertise two Node-SIDs (using SPF algorithm), one with the protection flag set, the other without the protection flag set. The same discussion regarding ECMP is also applicable here.
The remaining flag space in the Prefix-SID is small, so adding a new flag requires analysis but this should not be considered as a showstopper.
[I-D.ietf-spring-segment-routing] defines a Strict Shortest Path algorithm which mandates that the packet is forwarded according to ECMP-aware SPF algorithm and instructs any router in the path to ignore any possible local policy overriding SPF decision. The use of a local-repair for a strict SPF Node-SID is allowed as long as the FRR mechanism enforces the post convergence path to the destination.
This solution does not bring any benefit compared to the regular Node-SID (as it has similar properties).
Having two instances of the Node-SID (protected and not protected) is a requirement when using Node-SID in protected and non protected paths. The protection of a Node-SID is a matter of a local policy configuration on every node in the network. A service provider may configure two Node-SIDs per node and may adjust the local-repair on every node to protect one Node-SID but not the other. As the protection of the Node-SID is inherited from the protection of the associated prefix, the service provider will need to deploy a new set of prefixes to all nodes to deploy the new set of Node-SIDs. Then it will need to maintain the local-repair policy on every node to ensure that the prefixes associated to the non protected Node-SID are not using the local-repair.
The path computation engine (head-end or PCE) must be aware of the policy defined by the service provider so it can select the right SIDs/prefixes when computing a path.
One advantage of combining adjacency and node segments is the reduction of the label stack size.
The drawbacks are the increase of the controlplane and dataplane resource consumption. Whereas having two adjacency SIDs introduces a negligeable impact, having two nodes SIDs increases controlplane and dataplane processing as each node in the network will have to install an MPLS->MPLS and IP->MPLS entry for each additional Node-SID. The regular IP convergence time of the network may be doubled in the worst case while the newly deployed node-SIDs are only used for traffic-engineering applications. One of the other drawback is that a Node-SID may be transiently rerouted on a path that does not fit the constraints anymore if a transit node converges faster than the head-end: this concern is not new and applies to all traffic-engineering use cases. Note that there is a high chance for a transit node to reroute faster than the head-end as it has usually less computations to run (SPF+CSPFs) and less prefixes to rewrite; it may also run less features leaving more CPU slots for IGP reconvergence. The transient rerouting of the Node-SID may lead to microloops in the network that may impact the customer traffic. Node-SIDs are subject to ECMP and a local-repair mechanism may be implemented for equal cost paths with no way to disable it. If the requirement of preventing any local-repair or ECMP is strict, the path computation engine needs to prevent the usage of all Node-SIDs or needs to detect that a particular Node-SID will be subject to ECMP and enforce the usage of additional adjacency SIDs to break the ECMP. In any case, more adjacency-SIDs will be required in the stack to avoid the ECMP, leading to a deeper label stack.
[I-D.ietf-spring-segment-routing] defines the binding segment with multiple use cases. One of the use case of the binding segment is to advertise a tunnel as a segment. When a computation engine computes a non protected path and if the resulting label stack using only non protected adjacency segments is too deep for the network, an external component may create shortcuts in the network by creating a binding segment representing a list of non protected adjacency segments.
PE1--P1 P6 P10--PE2
\ / \ /
P4-P5 P7 P9
\ /
P8
Figure 3 - Use of Binding SID
In the example above, the path from PE1 to PE2 must be expressed with the stack: {Adj_P1P4,Adj_P4P5,Adj_P5P6,Adj_P6P7,Adj_P7P8,Adj_P8P9,Adj_P9P10,Adj_P10PE2}. This stack is too deep due to the limitations of the network. An external component may create a binding Binding1 on P5 that represents the non protected path (P5->P6->P7->P8->P9->P10). When the binding is created and advertised in the topology, the computation engine can use this binding SID in a path, resulting for a PE1 to PE2 path to the stack: {Adj_P1P4,Adj_P4P5,Binding1,Adj_P10PE2}. The usage of the binding SID in the stack allowed to reduce its size to an acceptable value.
One advantage of combining adjacency and binding segments is the reduction of the label stack size. The label stack size can be reduced to a small amount of labels at some price (creating some states on transit nodes).
The drawbacks are the increase of the controlplane and dataplane resource consumption. This controlplane and dataplane resource consumption are variable and will be linked to the intelligence of the external controller and computation engines and especially how the placement of the bindings is done to maximize the sharing between LSPs. Moreover any optimization try in the binding segment may introduce churn in the network controlplane (Make Before Break can be used to ensure that dataplane is not affected). Programming a binding-SID on a transit node is feasible only if the programming node has the necessary protocol sessions to do so. When a head-end router is performing a path computation, it is usually not the case. When a controller (PCE) is used, it may not have a session to all LSRs in the network, as only edge nodes may require a path computation. The controller may be limited for the placement of the binding SID to the nodes it has a protocol session with (it cannot setup a PCEP session by itself). A full deployment of protocol sessions with the controller may not be feasible for technical reasons (scaling, ...) or economical reasons. A potential mitigation could be to allow protocol sessions to be setup dynamically (when requirement comes) to an authorized subset of nodes in the network: some protocol modifications may be necessary to allow this behavior.
The following table tries to summarize the various solution pros/cons within a comparison table:
We consider a network with N nodes and L links, with an average of l links per node.
| Criteria | Solution 1 | Solution 2 | Solution 3 | Solution 4 | Solution 5 |
|---|---|---|---|---|---|
| Label stack size | One label per hop | Reduced | Reduced | Reduced | Reduced |
| Controlplane | Negligible | Potential additional computation + 2*N entries in RIB | + 2*N entries in RIB | + 2*N entries in RIB | Adds states in the LSRs |
| Dataplane | +l entries | +2*N entries | +2*N entries | +2*N entries | Variable |
| IP convergence time | None | Double | Double | Double | None |
| Computation engine | Needs to select Adj-SIDs with B=0 | Needs to select Adj-SIDs with B=0 and Node-SIDs with strict SPF | Needs to select Adj-SIDs with B=0 and Node-SIDs with B=0 | Needs to select Adj-SIDs with B=0 and needs to understand policy from the SP to select the right Node-SIDs | Needs to select Adj-SIDs with B=0 and place the binding SID in a smart way |
| Protocol | None | None | Need a new flag | None | None |
| ECMP avoidance | Supported | Supported at the price of increasing the label stack | Supported at the price of increasing the label stack | Supported at the price of increasing the label stack | Supported |
| Requirements fulfilment | Yes | Partially (allows ECMP+transient rerouting) | Partially (allows ECMP+transient rerouting) | Partially (allows ECMP+transient rerouting) | Yes |
| Others | None | None | None | None | Requires a controller with sessions to all nodes (even transit) |
Based on the analysis in Section 4, we only have two solutions that fulfill the requirements expressed in Section 2: usage of adjacency-SIDs only, usage of a combination of adjacency SIDs and binding SIDs.
As using only Adjacency-SIDs may reduce today the possibility of creating a path (due to the hardware/software limitations), authors would like to encourage the usage of a combination of adjacency-SIDs and binding-SIDs (Section 3.3) as a short-term solution.
However this approach has also several drawbacks, but authors think that these drawbacks can be reduced by enhancing existing protocols.
As a long term solution, authors would like to encourage vendors to support the ability for a node to push a significant number of labels, up to the full network diameter.
TBD.
Authors would like to thank Bruno Decraene for his valuable comments.
N/A
| [I-D.gredler-idr-bgp-ls-segment-routing-extension] | Gredler, H., Ray, S., Previdi, S., Filsfils, C., Chen, M. and J. Tantsura, "BGP Link-State extensions for Segment Routing", Internet-Draft draft-gredler-idr-bgp-ls-segment-routing-extension-02, October 2014. |
| [I-D.ietf-isis-segment-routing-extensions] | Previdi, S., Filsfils, C., Bashandy, A., Gredler, H., Litkowski, S., Decraene, B. and j. jefftant@gmail.com, "IS-IS Extensions for Segment Routing", Internet-Draft draft-ietf-isis-segment-routing-extensions-13, June 2017. |
| [I-D.ietf-ospf-segment-routing-extensions] | Psenak, P., Previdi, S., Filsfils, C., Gredler, H., Shakir, R., Henderickx, W. and J. Tantsura, "OSPF Extensions for Segment Routing", Internet-Draft draft-ietf-ospf-segment-routing-extensions-18, July 2017. |
| [I-D.ietf-spring-segment-routing] | Filsfils, C., Previdi, S., Decraene, B., Litkowski, S. and R. Shakir, "Segment Routing Architecture", Internet-Draft draft-ietf-spring-segment-routing-12, June 2017. |
| [RFC2119] | Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, DOI 10.17487/RFC2119, March 1997. |