Internet-Draft | LISP Traffic Engineering | May 2024 |
Farinacci, et al. | Expires 7 November 2024 | [Page] |
This document describes how LISP re-encapsulating tunnels can be used for Traffic Engineering purposes. The mechanisms described in this document require no LISP protocol changes but do introduce a new locator (RLOC) encoding. The Traffic Engineering features provided by these LISP mechanisms can span intra-domain, inter-domain, or combination of both.¶
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 7 November 2024.¶
Copyright (c) 2024 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 key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 [RFC2119][RFC8174] when, and only when, they appear in all capitals, as shown here.¶
This document describes extensions to the Locator/Identifier Separation Protocol (LISP) for traffic engineering features.¶
When LISP routers encapsulate packets to other LISP routers, the path stretch is typically 1, meaning the packet travels on a direct path from the encapsulating ITR to the decapsulating ETR at the destination site. The direct path is determined by the underlying routing protocol and metrics it uses to find the shortest path.¶
This specification will examine how re-encapsulating tunnels [RFC9300] can be used so a packet can take an administratively specified path, a congestion avoidance path, a failure recovery path, or multiple load-shared paths, as it travels from ITR to ETR. By introducing an Explicit Locator Path (ELP) locator encoding [RFC8060], an ITR can encapsulate a packet to a Re-Encapsulating Tunnel Router (RTR) which decapsulates the packet, then encapsulates it to the next locator in the ELP.¶
Refer to [RFC9300] for authoritative definitions for terms EID, RLOC, RTR, and Recursive Tunneling. The other terms defined in this section add to the canonical definition to reflect the design considerations in this specification.¶
Typically, a packet's path from source EID to destination EID travels through the locator core via the encapsulating ITR directly to the decapsulating ETR as the following diagram illustrates:¶
Legend:¶
Let's introduce RTRs 'X' and 'Y' so that, for example, if it is desirable to route around the path from B to C through link B-->C, one could provide an ELP of (X,Y,etr):¶
There are various reasons why the path from 'seid' to 'deid' may want to avoid the path from B to C. To list a few:¶
There may not be sufficient capacity provided by the networks that connect B and C together.¶
There may be a policy reason to avoid the ASes that make up the path between B and C.¶
There may be a failure on the path between B and C which makes the path unreliable.¶
There may be monitoring or traffic inspection resources close to RTRs X and Y that do network accounting or measurement.¶
There may be a chain of services performed at RTRs X and Y regardless if the path from ITR to ETR is through B and C.¶
The notation for a general formatted ELP is (x, y, etr) which represents the list of RTRs a packet SHOULD travel through to reach the final tunnel hop to the ETR.¶
The procedure for using an ELP at each tunnel hop is as follows:¶
The ITR will retrieve the ELP from the mapping database.¶
The ITR will encapsulate the packet to RLOC 'x'.¶
The RTR with RLOC 'x' will decapsulate the packet. It will use the decapsulated packet's destination address as a lookup into the mapping database to retrieve the ELP.¶
RTR 'x' will encapsulate the packet to RTR with RLOC 'y'.¶
The RTR with RLOC 'y' will decapsulate the packet. It will use the decapsulated packet's destination address as a lookup into the mapping database to retrieve the ELP.¶
RTR 'y' will encapsulate the packet on the final tunnel hop to ETR with RLOC 'etr'.¶
The ETR will decapsulate the packet and deliver the packet to the EID inside of its site.¶
The specific encoding format for the ELP can be found in [RFC8060]. It is defined that an ELP will appear as a single encoded locator in a locator-set. Say for instance, we have a mapping entry for EID-prefix 10.0.0.0/8 that is reachable via 4 locators. Two locators are being used as active/active and the other two are used as active/active if the first two go unreachable (as noted by the priority assignments below). This is what the mapping entry would look like:¶
EID-prefix: 10.0.0.0/8 Locator-set: ETR-A: priority 1, weight 50 ETR-B: priority 1, weight 50 ETR-C: priority 2, weight 50 ETR-D: priority 2, weight 50¶
If an ELP is going to be used to have a policy path to ETR-A and possibly another policy path to ETR-B, the locator-set would be encoded as follows:¶
EID-prefix: 10.0.0.0/8 Locator-set: (x, y, ETR-A): priority 1, weight 50 (q, r, ETR-B): priority 1, weight 50 ETR-C: priority 2, weight 50 ETR-D: priority 2, weight 50¶
The mapping entry with ELP locators is registered to the mapping database system, see [RFC9301] for details, just like any other mapping entry would. The registration is typically performed by the ETR(s) that are assigned and own the EID-prefix. That is, the destination site makes the choice of the RTRs in the ELP. Alternatively, it may be common practice for a third-party system (not an ETR network entity) to register ELP mappings. This can be done via a general purpose SDN provisioning system, for example.¶
Another case where a locator-set can be used for flow-based load-sharing across multiple paths to the same destination site:¶
EID-prefix: 10.0.0.0/8 Locator-set: (x, y, ETR-A): priority 1, weight 75 (q, r, ETR-A): priority 1, weight 25¶
Using this mapping entry, an ITR would load split 75% of the EID flows on the (x, y, ETR-A) ELP path and 25% of the EID flows on the (q, r, ETR-A) ELP path. If any of the ELPs go down, then the other can take 100% of the load. For mapping system lookups and map-cache management, see [RFC9300] for details.¶
ELP re-optimization is a process of changing the RLOCs of an ELP due to underlying network change conditions. Just like when there is any locator change for a locator-set, the procedures from the main LISP specification [RFC9300] are followed.¶
When a RLOC from an ELP is changed, Map-Notify messages [RFC9301] can be used to inform the existing RTRs in the ELP so they can do a lookup to obtain the latest version of the ELP. Map-Notify messages can also be sent to new RTRs in an ELP so they can get the ELP in advance to receiving packets that will use the ELP. This can minimize packet loss during mapping database lookups in RTRs.¶
In the previous examples, we showed how an ITR encapsulates using an ELP of (x, y, etr). When a packet is encapsulated by the ITR to RTR 'x', the RTR may want a policy path to RTR 'y' and run another level of re-encapsulating tunnels for packets destined to RTR 'y'. In this case, RTR 'x' does not encapsulate packets to 'y' but rather performs a mapping database lookup on the address 'y', which returns a ELP-based locator record for a path to RTR 'y', and encapsulates packets to the first-hop of the returned ELP. If the ELP path to RTR 'y' is an internal path within a LISP site, the lookup for RTR 'y' can be done to a private mapping system. The decision to use address 'y' as an encapsulation address versus a lookup address is based on the L-bit setting for 'y' in the ELP entry. The decision and policy of ELP encodings are local to the entity which registers the EID-prefix associated with the ELP.¶
Another example of recursion is when the ITR uses the ELP (x, y, etr) to first prepend a header with a destination RLOC of the ETR and then prepend another header and encapsulate the packet to RTR 'x'. When RTR 'x' decapsulates the packet, rather than doing a mapping database lookup on RTR 'y' the last example showed, instead RTR 'x' does a mapping database lookup on ETR 'etr'. In this scenario, RTR 'x' can choose an ELP from the locator-set by considering the source RLOC address of the ITR versus considering the source EID.¶
This additional level of recursion also brings advantages for the provider of RTR 'x' to store less state. Since RTR 'x' does not need to look at the inner most header, it does not need to store EID state. It only stores an entry for RTR 'y' which many EID flows could share for scaling benefits. The locator-set for entry 'y' could either be a list of typical locators, a list of ELPs, or combination of both. Another advantage is that packet load-splitting can be accomplished by examining the source of a packet. If the source is an ITR versus the source being the last-hop of an ELP the last-hop selected, different forwarding paths can be used.¶
Paths to an ETR may want to be selected based on different classes of service. Packets from a set of sources that have premium service can use ELP paths that are less congested where normal sources use ELP paths that compete for less resources or use longer paths for best effort service.¶
Using source/destination lookups into the mapping database can yield different ELPs. For example, a premium service flow with (source=1.1.1.1, dest=10.1.1.1) can be described by using the following mapping entry:¶
EID-prefix: (1.0.0.0/8, 10.0.0.0/8) Locator-set: (x, y, ETR-A): priority 1, weight 50 (q, r, ETR-A): priority 1, weight 50¶
And all other best-effort sources would use different mapping entry described by:¶
EID-prefix: (0.0.0.0/0, 10.0.0.0/8) Locator-set: (x, x', y, y', ETR-A): priority 1, weight 50 (q, q', r, r', ETR-A): priority 1, weight 50¶
If the source/destination lookup is coupled with recursive lookups, then an ITR can encapsulate to the ETR, prepending a header that selects source address ITR-1 based on the premium class of service source, or selects source address ITR-2 for best-effort sources with normal class of service. The ITR then does another lookup in the mapping database on the prepended header using lookup key (source=ITR-1, dest=10.1.1.1) that returns the following mapping entry:¶
EID-prefix: (ITR-1, 10.0.0.0/8) Locator-set: (x, y, ETR-A): priority 1, weight 50 (q, r, ETR-A): priority 1, weight 50¶
And all other sources would use different mapping entry with a lookup key of (source=ITR-2, dest=10.1.1.1):¶
EID-prefix: (ITR-2, 10.0.0.0/8) Locator-set: (x, x', y, y', ETR-A): priority 1, weight 50 (q, q', r, r', ETR-A): priority 1, weight 50¶
This will scale the mapping system better by having fewer source/destination combinations. Refer to the Source/Dest LCAF type described in [RFC8060] for encoding EIDs in Map-Request and Map-Register messages.¶
An ELP that is first used by an ITR must be inspected for encoding loops. If any RLOC appears twice in the ELP, it MUST not be used.¶
Since it is expected that multiple mapping systems will be used, there can be a loop across ELPs when registered in different mapping systems. The TTL copying procedures for re-encapsulating tunnels and recursive tunnels in [RFC9300] MUST be followed.¶
An ELP can be used to deploy services at each reencapsulation point in the network. One example is to implement a honey-pot service when a destination EID is being DoS attacked. That is, when a DoS attack is recognized when the encapsulation path is between ITR and ETR, an ELP can be registered for a destination EID to the mapping database system. The ELP can include an RTR so the ITR can encapsulate packets to the RTR which will decapsulate and deliver packets to a scrubber service device. The scrubber could decide if the offending packets are dropped or allowed to be sent to the destination EID. In which case, the scurbber delivers packets back to the RTR which encapsulates to the ETR.¶
Since an RTR knows the next tunnel hop to encapsulate to, it can monitor the reachability of the next-hop RTR RLOC by doing RLOC-probing according to the procedures in [RFC9300]. When the RLOC is determined unreachable by the RLOC-probing mechanisms, the RTR can use another locator in the locator-set. That could be the final ETR, a RLOC of another RTR, or an ELP where it must search for itself and use the next RLOC in the ELP list to encapsulate to.¶
RLOC-probing can also be used to measure delay on the path between RTRs and when it is desirable switch to another lower delay ELP.¶
Since an ELP-node knows the reachabiliy of the next ELP-node in a ELP by using RLOC probing, the sum of reachability can determine the reachability of the entire path. A head-end ITR/RTR/PITR can determine the quality of a path and decide to select one path from another based on the telemetry data gathered by RLOC-probing for each encapsulation hop.¶
ELP-probing mechanism details can be found in [I-D.filyurin-lisp-elp-probing].¶
[RFC6832] defines procedures for how non-LISP sites talk to LISP sites. The network elements defined in the Interworking specification, the proxy ITR (PITR) and proxy ETR (PETR) (as well as their multicast counterparts defined in [RFC6831]) can participate in LISP-TE. That is, a PITR and a PETR can appear in an ELP list and act as an RTR.¶
Note when an RLOC appears in an ELP, it can be of any address-family. There can be a mix of IPv4 and IPv6 locators present in the same ELP. This can provide benefits where islands of one address-family or the other are supported and connectivity across them is necessary. For instance, an ELP can look like:¶
(x4, a46, b64, y4, etr)¶
Where an IPv4 ITR will encapsulate using an IPv4 RLOC 'x4' and 'x4' could reach an IPv4 RLOC 'a46', but RTR 'a46' encapsulates to an IPv6 RLOC 'b64' when the network between them is IPv6-only. Then RTR 'b64' encapsulates to IPv4 RLOC 'y4' if the network between them is dual-stack.¶
Note that RTRs can be used for NAT-traversal scenarios [I-D.ermagan-lisp-nat-traversal] as well to reduce the state in both an xTR that resides behind a NAT and the state the NAT needs to maintain. In this case, the xTR only needs a default map-cache entry pointing to the RTR for outbound traffic and all remote ITRs can reach EIDs through the xTR behind a NAT via a single RTR (or a small set RTRs for redundancy).¶
RTRs have some scaling features to reduce the number of locator-set changes, the amount of state, and control packet overhead:¶
When ITRs and PITRs are using a small set of RTRs for encapsulating to "orders of magnitude" more EID-prefixes, the probability of locator-set changes are limited to the RTR RLOC changes versus the RLOC changes for the ETRs associated with the EID-prefixes if the ITRs and PITRs were directly encapsulating to the ETRs. This comes at an expense in packet stretch, but depending on RTR placement, this expense can be mitigated.¶
When RTRs are on-path between many pairwise EID flows, ITRs and PITRs can store a small number of coarse EID-prefixes.¶
RTRs can be used to help scale RLOC-probing. Instead of ITRs RLOC-probing all ETRs for each destination site it has cached, the ITRs can probe a smaller set of RTRs which in turn, probe the destination sites.¶
ELPs have application in multicast environments. Just like RTRs can be used to provide connectivity across different address family islands, RTRs can help concatenate a multicast region of the network to one that does not support native multicast.¶
Note there are various combinations of connectivity that can be accomplished with the deployment of RTRs and ELPs:¶
Providing multicast forwarding between IPv4-only-unicast regions and IPv4-multicast regions.¶
Providing multicast forwarding between IPv6-only-unicast regions and IPv6-multicast regions.¶
Providing multicast forwarding between IPv4-only-unicast regions and IPv6-multicast regions.¶
Providing multicast forwarding between IPv6-only-unicast regions and IPv4-multicast regions.¶
Providing multicast forwarding between IPv4-multicast regions and IPv6-multicast regions.¶
An ITR or PITR can do a (S-EID,G) lookup into the mapping database. What can be returned is a typical locator-set that could be made up of the various RLOC addresses:¶
The locator-set above can be used as a replication list. That is some RLOCs listed can be unicast RLOCs and some can be delivery group RLOCs. A unicast RLOC in this case is used to encapsulate a multicast packet originated by a multicast source EID into a unicast packet for unicast delivery on the underlying network. ETR-A could be an IPv4 unicast RLOC address and ETR-B could be a IPv6 unicast RLOC address.¶
A delivery group address is used when a multicast packet originated by a multicast source EID is encapsulated in a multicast packet for multicast delivery on the underlying network. Group address 'g1' could be an IPv4 delivery group RLOC and group address 'g2' could be an IPv6 delivery group RLOC.¶
Flexibility for these various types of connectivity combinations can be achieved and provided by the mapping database system. And the RTR placement allows the connectivity to occur where the differences in network functionality is located.¶
Extending this concept by allowing ELPs in locator-sets, one could have this locator-set registered in the mapping database for (S-EID, G). For example:¶
In the above situation, an ITR would encapsulate a multicast packet originated by a multicast source EID to the RTR with unicast RLOC 'x'. Then RTR 'x' would decapsulate and unicast encapsulate to RTR 'y' ('x' or 'y' could be either IPv4 or IPv6 unicast RLOCs), which would decapsulate and unicast encapsulate to the final RLOC 'ETR-A'. The ETR 'ETR-A' would decapsulate and deliver the multicast packet natively to all the receivers joined to application group 'G' inside the LISP site.¶
Let's look at the ITR using the ELP (a, g, b, ETR-B). Here the encapsulation path would be the ITR unicast encapsulates to unicast RLOC 'a'. RTR 'a' multicast encapsulates to delivery group 'g'. The packet gets to all ETRs that have joined delivery group 'g' so they can deliver the multicast packet to joined receivers of application group 'G' in their sites. RTR 'b' is also joined to delivery group 'g'. Since it is in the ELP, it will be the only RTR that unicast encapsulates the multicast packet to ETR 'ETR-B'. Lastly, 'ETR-B' decapsulates and delivers the multicast packet to joined receivers to application group 'G' in its LISP site.¶
As one can see there are all sorts of opportunities to provide multicast connectivity across a network with non-congruent support for multicast and different address-families. One can also see how using the mapping database can allow flexible forms of delivery policy, rerouting, and congestion control management in multicast environments.¶
When an RTR receives a LISP encapsulated packet, it can look at the outer source address to verify that RLOC is the one listed as the previous hop in the ELP list. If the outer source RLOC address appears before the RLOC which matches the outer destination RLOC address, the decapsulating RTR (or ETR if last hop), SHOULD choose to drop the packet.¶
This document does not make any request to IANA.¶
The authors would like to thank the following people for their ideas and comments. They are Albert Cabellos, Khalid Raza, and Vina Ermagan, Gregg Schudel, Yan Filyurin, Robert Raszuk, and Truman Boyes.¶
Posted July 2012.¶
Add the Lookup bit to allow an ELP to be a list of encapsulation and/or mapping database lookup addresses.¶
Indicate that ELPs can be used for service chaining.¶
Add text to indicate that Map-Notify messages can be sent to new RTRs in a ELP so their map-caches can be pre-populated to avoid mapping database lookup packet loss.¶
Fixes to editorial comments from Gregg.¶
Initial draft posted March 2012.¶