Network Working Group D. Saucez (Ed.) Internet-Draft INRIA Intended status: Informational October 15, 2012 Expires: April 18, 2013 LISP deployment impact draft-saucez-lisp-impact-00.txt Abstract The Locator/Identifier Separation Protocol (LISP) relies on three simple principles to scale the Internet: address role separation, encapsulation, and mapping. In this document, based on implementation, deployment, and theoretical studies, we discuss the impact that a LISP deployment would have on the Internet and for the users. Status of this Memo 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 April 18, 2013. Copyright Notice Copyright (c) 2012 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 Saucez (Ed.) Expires April 18, 2013 [Page 1] Internet-Draft LISP impact October 2012 described in the Simplified BSD License. Table of Contents 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 3 2. LISP in a nutshell . . . . . . . . . . . . . . . . . . . . . . 3 3. LISP for scaling the Internet . . . . . . . . . . . . . . . . 4 4. Beyond scaling the Internet . . . . . . . . . . . . . . . . . 5 4.1. Traffic engineering . . . . . . . . . . . . . . . . . . . 5 4.2. IPv4/IPv6 Transition . . . . . . . . . . . . . . . . . . . 6 4.3. Inter-domain multicast . . . . . . . . . . . . . . . . . . 6 5. Impact of LISP on operations and business model . . . . . . . 7 5.1. Impact on non-LISP traffic and sites . . . . . . . . . . . 7 5.2. Impact on LISP traffic and sites . . . . . . . . . . . . . 8 6. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 9 7. Security Considerations . . . . . . . . . . . . . . . . . . . 9 8. Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . 9 9. References . . . . . . . . . . . . . . . . . . . . . . . . . . 9 9.1. Normative References . . . . . . . . . . . . . . . . . . . 9 9.2. Informative References . . . . . . . . . . . . . . . . . . 11 Author's Address . . . . . . . . . . . . . . . . . . . . . . . . . 12 Saucez (Ed.) Expires April 18, 2013 [Page 2] Internet-Draft LISP impact October 2012 1. Introduction The Locator/Identifier Separation Protocol (LISP) relies on three simple principles to scale the Internet: address role separation, encapsulation, and mapping. The main target of LISP is to make the Internet more scalable by reducing the number of routes in the DFZ as well as the churn. As LISP relies on mapping and encapsulation, it offers more than just scalability. For example, LISP provides a mean for a LISP site to precisely control is inter-domain outgoing and incoming traffic, with the possibility to apply different policies to the different domains exchanging traffic with it. LISP can also be used during the IPv/IPv6 transition phase as it allows to transport IPv4 over IPv6 or IPv6 over IPv4. LISP also provides a solution to perform inter-domain multicast. In this document we discuss the impact a LISP deployment would have on the Internet and for the users. We first show that interworking results in path stretch and there is still open questions related to the deployment model of such interworking (not technical but economical). Afterward, we show that encapsulation causes issue (but not in practice) because it reduces the MTU. An important impact of LISP on operation is related to resiliency and troubleshooting. Indeed, as LISP relies on mappings that are cached and on encapsulation, troubleshooting is harder than in the traditional Internet. Also, end-to-end encapsulation also stress resiliency as it makes failure detection and recovery slower than with hop-by-hop routing. 2. LISP in a nutshell The Locator/Identifier Separation Protocol (LISP) relies on three simple principles: address role separation, encapsulation, and mapping. Addresses are separated in two roles: the Routing Locators (RLOCs) and the Endpoint Identifiers (EIDs). RLOCs are assigned from the address space of the Internet service providers. The EIDs are attributed by block extracted from the EID Space. To limit the scalability problem of today's Internet, only the routes towards the RLOCs are announced on the Internet while EIDs are also propagated today. LISP routers are used at the boundary between the EID and the RLOC spaces. Routers used to exit the EID space are called Ingress Tunnel Router (ITRs) and those used to enter the EID space the Egress Tunnel Routers (ETRs). When a host sends a packet to a remote destination, it sends it as in today's Internet. The packet eventually arrives at Saucez (Ed.) Expires April 18, 2013 [Page 3] Internet-Draft LISP impact October 2012 the border of its site at an ITR. Because EIDs are not routable on the Internet, the packet is encapsulated with the source address set to the ITR RLOC and the destination address set to the ETR RLOC. The encapsulated packet is then forwarded in the Internet until it reaches the selected ETR. The ETR decapsulates the packet and forwards it to the destination. The acronym xTR for Ingress/Egress tunnel router is used for a router playing these two roles. The correspondence between EIDs and RLOCs is given by the mappings. When an ITR needs to find ETR RLOCs that serve an EID it queries the mapping system. It is worth noticing that LISP is not restricted to the Internet Protocol for the EID addresses. Indeed, the LISP Canonical Address Format (LCAF) allows encoding any kind of address [I-D.farinacci-lisp-lcaf]. Therefore, any address type can be used as EID (the address is the key for the mapping lookup) and LISP can then transport, for example, Ethernet frames over the Internet. A more throughout introduction to LISP can be found in [I-D.chiappa-lisp-introduction] and a discussion around the architecture in [I-D.chiappa-lisp-architecture]. The complete specifications are given in [I-D.ietf-lisp], [I-D.ietf-lisp-ms], [I-D.fuller-lisp-ddt], [I-D.ietf-lisp-interworking], [I-D.ietf-lisp-map-versioning], and [I-D.ietf-lisp-sec]. 3. LISP for scaling the Internet The first goal of LISP is to scale the Internet. LISP improves the Internet scalability because traffic engineering and stub do not appear in the core anymore, so smaller and more stable tables in the core. Also, at the edge, the routing tables scale with the edge, its own, traffic pattern and not the size of the Internet. Scaling improvement is proven by several works. Quoitin et al. show in [QIdLB07] that the locator/identifier separation improves the routing scalability by reducing the FIB size and increases the path diversity and thus the traffic engineering capabilities. In addition, Iannone and Bonaventure show in [IB07] that the number of mapping entries that must be supported at an ITR of a campus network is limited and does not represent more that 3 to 4 Megabytes of memory. Similarly, Juhoon et al. show that the EID- to-RLOC Cache size should not exceed 14 MB for an ITR responsible of more than 20,000 residential ADSL users at a large ISP [KIF11]. [IB07], [KIF11] rely on BGP and traffic traces to determine the number of entries to keep in the cache. In both papers, the size of the cache is inferred from the number of entries by considering that every EID is associated with two or three locators. [S11] confirms these results by looking at the distribution of the number of Saucez (Ed.) Expires April 18, 2013 [Page 4] Internet-Draft LISP impact October 2012 locators per EID if LISP were deployed in today's Internet. The assumptions in these studies are: o contiguous addresses tend to be used similarly, EID prefixes follow the current BGP prefixes decomposition; o EIDs are used only at the stub ASes, not in the transit ASes; o the RLOCs of an EID are deployed at the edge between the stubs owning the EID and the providers and locator addresses are allocated in a Provider Aggregetable (PA) mode. 4. Beyond scaling the Internet Even though it is its main goal, LISP is more than just a scalability solution, it is also a tool to provide both incoming and outgoing traffic engineering [S11], can be used as an IPv6 transition at the routing level, and inter-domain multicast [I-D.ietf-lisp-multicast], [I-D.coras-lisp-re]. LISP has also proven to be a good protocol for mobility of devices in the Internet [I-D.meyer-lisp-mn] or even virtual machine mobility in data centers and multi-tenant VPN, however, we will not discuss here these two last points as they are out of the scope of the charter. 4.1. Traffic engineering In today's Internet, stub networks are globally routable and the routing system distributes the routes to reach these stubs. On the contrary, the EID prefixes of a LISP site are not routable on the Internet and mappings are needed to determine the list of LISP routers to contact to send them packets. The difference is significant for two reasons. First, packets are not sent to a site but to a specific ingress router. Second, a site can control the entry points for its traffic by controlling its mappings. For traffic engineering purpose, a mapping associate an EID prefix to a list of RLOCs. Each RLOC is annotated with a priority and a weight. When there is several RLOCs, the ITR selects the one with the lowest priority value and sends the encapsulated packet to this RLOC. If several such RLOCs exist, then the traffic is balanced proportionally to the weight among the different RLOC with the lowest priority value. Traffic engineering in LISP thus allows the mapping owner to have a fine-grained control on the primary and backup path its incoming and outgoing packet use. In addition, it can share the load among its links. Traffic engineering in LISP goes one step further. Indeed, every Saucez (Ed.) Expires April 18, 2013 [Page 5] Internet-Draft LISP impact October 2012 Map-Request contains the Source EID Address of the packet that caused a cache miss and triggered the Map-Request. It is thus possible for a mapping owner to differentiate the answer (Map-Reply) it gives to Map-Requests based on the requester. This functionality is not possible today because a domain cannot control exactly the routes that will be received by domains that are not in the direct neighborhood. 4.2. IPv4/IPv6 Transition The LISP encapsulation mechanism is designed to support any combination of locators and identifiers address family. It is then possible to bind IPv6 EIDs with IPv4 RLOCs and vice-versa. This allows transporting IPv6 packets over an IPv4 network (or IPv4 packets over an IPv6 network), thus enabling the use of LISP as an IPv6 transition mechanism. A not so uncommon example is the case of the network infrastructure of a datacenter being IPv4-only while dual-stack front-end load balancers are used. In this scenario, LISP can be used to provide IPv6 access to servers even though the network and the servers only support IPv4. Indeed, assuming that the datacenter's ISP offers IPv6 connectivity, the datacenter only needs to deploy one (or more) xTR(s) at its border with the ISP and one (or more) xTR(s) directly connected to the load balancers. The xTR(s) at the ISP's border tunnels IPv6 packets over IPv4 to the xTR(s) directly attached to the load balancer. The load balancer's xTRs decapsulate the packets and forward them to the load balancers, which act as proxies, translating each IPv6 packet into an IPv4. IPv4 packets are then sent to the appropriate servers. Similarly, when the server response arrives at the load balancer, the packet is translated back into an IPv6 packet and forwarded to its xTR(s), which in turn will tunnel it back, over the IPv4-only infrastructure, to the xTR(s) connected to the ISP. The packet is then decapsulated and forwarded to the ISP natively in IPv6. 4.3. Inter-domain multicast LISP naturally LISP supports multicast as a list of locators is associated to the prefixes, it is thus just necessary to duplicate the packets to multiple RLOCs instead of selecting only one [I-D.ietf-lisp-multicast]. [I-D.coras-lisp-re] and [CDM+12] propose a technique to construct multicast distribution tree and the test of three different management strategies for low latency content delivery show that such overlay can support thousands of members and hundreds of thousands of clients. Interestingly, when constructed wisely, high client churn Saucez (Ed.) Expires April 18, 2013 [Page 6] Internet-Draft LISP impact October 2012 has a limited impact on the performance and management overhead. 5. Impact of LISP on operations and business model Important implementation efforts (Cisco NXOS, Cisco IOS, FreeBSD OpenLISP, Linux LISPMob, LISP-Click) have been made to assess the specifications and interoperability tests [Was09] have been a success. Large scale deployment in the international lisp4.net testbed permit to learn operational matters related to LISP. We have to distinguish the impact on LISP sites and traffic of the impact on non-LISP site and traffic. 5.1. Impact on non-LISP traffic and sites LISP has no impact on traffic which has neither LISP origin nor LISP destination. However, LISP can have a significant impact on traffic between a LISP site and a non-LISP site. Indeed, traffic between a non-LISP site and a LISP site are subject to the same issues than those observed for LISP-to-LISP traffic (cf supra) but also have issues specific to the transition mechanism that allow LISP site to exchange packets with non-LISP site ([I-D.ietf-lisp-interworking], [I-D.ietf-lisp-deployment]). Indeed, the transition requires proxies tunnel routers (PxTRs). PxTRs do not cause particular technical issue. However, by definition proxies cause path stretch and make troubleshooting harder. There are still big questions related to PxTRs that have to be answered: o Where to deploy PxTRs? The placement in the topology has an important impact on the path stretch. o How many PxTRs? The number of PxTR has a direct impact on there load and the impact of the failure of a PxTR on the traffic. o What part of the EID space? Will all the PxTRs be proxies for the whole EID space or will it be segmented between different PxTRs? o Who to operate PxTRs? The IETF does not aim at providing business model hints, however, an important question to answer is related to the entities that will deploy PxTRs, how they will earn money and how the traffic will be carried with respect for the security and privacy. PxTR also normally have to advertise the EID prefix they are proxy for in BGP. However, if proxies are managed by different entities, Saucez (Ed.) Expires April 18, 2013 [Page 7] Internet-Draft LISP impact October 2012 they will belong to different ASes. In this case, we have to be sure that it will not cause MOA issues that could negatively influence routing. 5.2. Impact on LISP traffic and sites LISP is a protocol based on the map-and-encap paradigm which has the positive effects that we have given in the sections above. However, by construction, LISP also have side impact on operations: o MTU issue: as LISP uses encapsulation, the MTU is reduced (by 36 bytes in IPv4), this has implication on potentially all the traffic. However, in practice, on the lisp4.net network, we have not seen major issue due to the MTU. This is probably due to the fact that current end-host stacks are well designed to deal with the problem of MTU. o Resiliency issue: the advantage of flexibility and control offered by the loc/ID separation comes at the cost of increasing the complexity of the reachability detection. Indeed, identifiers are not directly routable and have to be mapped to locators. But a locator may be unreachable while others are still reachable. This is an important problem for any tunnel-based solution. In the current Internet, packets are forwarded independently of the border router of the network meaning that in case of the failure of a border router, another one can be used. With LISP, the destination RLOC specifically designate one particular ETR, hence if this ETR fails, the traffic is dropped even though other ETRs are available for the destination site. Another resiliency issue is linked to the fact that mappings are learned on demand. When an ITR fails, all its traffic is redirected to other ITRs that might not have yet the mappings for the redirected traffic. The study in [SKI+12] and [SD12] show, based on measurements and traffic traces, that failure of ITRs and RLOC are seldom but that when such failure happens, an important number of packet can be dropped. Unfortunately, the current techniques for LISP resiliency, based on monitoring or probing are not rapid enough (failure recovery of the order of a few seconds). To tackle this issue [I-D.bonaventure-lisp-preserve] and [I-D.saucez-lisp-itr-graceful] propose techniques based on local failure detection and recovery. o Middle boxes/filters: because of encapsulation, the middle boxes might not understand the traffic which can cause firewall to drop legitimate packets. In addition, LISP allows triangular or even rectangular routing, so it is hard to maintain a correct state even if the middle box perfectly understands LISP. Finally, filtering might also have problems because they might think only Saucez (Ed.) Expires April 18, 2013 [Page 8] Internet-Draft LISP impact October 2012 one host is generating the traffic (the ITR), as long as it is not decapsulated. o Troubleshooting/debugging: the major issue years of LISP experimentation have shown is the difficulty of troubleshooting. When there is a problem in the network, it is hard to pin-point the reason as the operator only has an inconsistent view of the network. He sees what is in its cache/database, he can try to see what is potentially elsewhere by querying the MRs but the knowledge is partial. On top of that, ICMP is too small, which means that when an ICMP arrives the ITR, it might not contain enough information to troubleshoot correctly. Furthermore, deployment in the beta network has shown that LISP+ALT was not easy to maintain and control, hence the migration to LISP-DDT. o Business: the IETF is not aiming at providing business models. However, even though [IL10] shown that there is economical incentives to migrate to LISP, some questions are on hold. For example, how will the EID be allocated to allow aggregation and hence scalability of the mapping system? Who will operate the mapping system infrastructure and for what benefit? 6. IANA Considerations TBD 7. Security Considerations As every protocol, LISP is subject to security issues. A detailed analysis of LISP security threats is given in [I-D.ietf-lisp-threats]. 8. Acknowledgments The people that contributed to this document are: Florin Coras, Vince Fuller, Joel Halpern, Luigi Iannone, Terry Manderson, Fabio Maino, and Gregg Schudel. 9. References 9.1. Normative References [I-D.fuller-lisp-ddt] Fuller, V., Lewis, D., Ermagan, V., and A. Jain, "LISP Saucez (Ed.) Expires April 18, 2013 [Page 9] Internet-Draft LISP impact October 2012 Delegated Database Tree", draft-fuller-lisp-ddt-04 (work in progress), September 2012. [I-D.ietf-lisp] Farinacci, D., Fuller, V., Meyer, D., and D. Lewis, "Locator/ID Separation Protocol (LISP)", draft-ietf-lisp-23 (work in progress), May 2012. [I-D.ietf-lisp-alt] Fuller, V., Farinacci, D., Meyer, D., and D. Lewis, "LISP Alternative Topology (LISP+ALT)", draft-ietf-lisp-alt-10 (work in progress), December 2011. [I-D.ietf-lisp-deployment] Jakab, L., Cabellos-Aparicio, A., Coras, F., Domingo- Pascual, J., and D. Lewis, "LISP Network Element Deployment Considerations", draft-ietf-lisp-deployment-04 (work in progress), September 2012. [I-D.ietf-lisp-interworking] Lewis, D., Meyer, D., Farinacci, D., and V. Fuller, "Interworking LISP with IPv4 and IPv6", draft-ietf-lisp-interworking-06 (work in progress), March 2012. [I-D.ietf-lisp-map-versioning] Iannone, L., Saucez, D., and O. Bonaventure, "LISP Map- Versioning", draft-ietf-lisp-map-versioning-09 (work in progress), March 2012. [I-D.ietf-lisp-ms] Fuller, V. and D. Farinacci, "LISP Map Server Interface", draft-ietf-lisp-ms-16 (work in progress), March 2012. [I-D.ietf-lisp-multicast] Farinacci, D., Meyer, D., Zwiebel, J., and S. Venaas, "LISP for Multicast Environments", draft-ietf-lisp-multicast-14 (work in progress), February 2012. [I-D.ietf-lisp-sec] Maino, F., Ermagan, V., Cabellos-Aparicio, A., Saucez, D., and O. Bonaventure, "LISP-Security (LISP-SEC)", draft-ietf-lisp-sec-04 (work in progress), October 2012. Saucez (Ed.) Expires April 18, 2013 [Page 10] Internet-Draft LISP impact October 2012 9.2. Informative References [CDM+12] Coras, F., Domingo-Pascual, J., Maino, F., Farinacci, D., and A. Cabellos-Aparicio, "Lcast: Software-defined Inter- Domain Multicast", Technical Report, Universitat Politecnica de Catalunya, 2012. [I-D.bonaventure-lisp-preserve] Bonaventure, O., Francois, P., and D. Saucez, "Preserving the reachability of LISP ETRs in case of failures", draft-bonaventure-lisp-preserve-00 (work in progress), July 2009. [I-D.chiappa-lisp-architecture] Art, Y., "An Architectural Perspective on the LISP Location-Identity Separation System", draft-chiappa-lisp-architecture-01 (work in progress), July 2012. [I-D.chiappa-lisp-introduction] Art, Y., "An Introduction to the LISP Location-Identity Separation System", draft-chiappa-lisp-introduction-01 (work in progress), July 2012. [I-D.coras-lisp-re] Coras, F., Cabellos-Aparicio, A., Domingo-Pascual, J., Maino, F., and D. Farinacci, "LISP Replication Engineering", draft-coras-lisp-re-00 (work in progress), July 2012. [I-D.farinacci-lisp-lcaf] Farinacci, D., Meyer, D., and J. Snijders, "LISP Canonical Address Format (LCAF)", draft-farinacci-lisp-lcaf-10 (work in progress), July 2012. [I-D.ietf-lisp-threats] Saucez, D., Iannone, L., and O. Bonaventure, "LISP Threats Analysis", draft-ietf-lisp-threats-02 (work in progress), September 2012. [I-D.meyer-lisp-mn] Farinacci, D., Lewis, D., Meyer, D., and C. White, "LISP Mobile Node", draft-meyer-lisp-mn-07 (work in progress), April 2012. [I-D.saucez-lisp-itr-graceful] Saucez, D., Bonaventure, O., Iannone, L., and C. Filsfils, "LISP ITR Graceful Restart", Saucez (Ed.) Expires April 18, 2013 [Page 11] Internet-Draft LISP impact October 2012 draft-saucez-lisp-itr-graceful-00 (work in progress), July 2012. [IB07] Iannone, L. and O. Bonaventure, "On the cost of caching locator/id mappings", In Proc. ACM CoNEXT 2007. [IL10] Iannone, L. and T. Leva, "Modeling the economics of Loc/ID Separation for the Future Internet", Book Chapter, Towards the Future Internet - Emerging Trends from the European Research, IOS Press. [KIF11] Kim, J., Iannone, L., and A. Feldmann, "Deep dive into the lisp cache and what isps should know about it", In Proc. IFIP Networking 2011. [QIdLB07] Quoitin, B., Iannone, L., de Launois, C., and O. Bonaventure, "Evaluating the benefits of the locator/ identifier separation", In Proc. ACM MobiArch 2007. [S11] Saucez, D., "Mechanisms for Interdomain Traffic Engineering with LISP", PhD Thesis, Universite catholique de Louvain, 2011. [SD12] Saucez, D. and B. Donnet, "On the Dynamics of Locators in LISP", In Proc. IFIP Networking 2012. [SKI+12] Saucez, D., Kim, J., Iannone, L., Bonaventure, O., and C. Filsfils, "A Local Approach to Fast Failure Recovery of LISP Ingress Tunnel Routers", In Proc. IFIP Networking 2012. [Was09] Wasserman, M., "LISP Interoperability Testing", IETF 76, LISP WG presentation, 2009. Author's Address Damien Saucez INRIA 2004 route des Lucioles BP 93 06902 Sophia Antipolis Cedex France Email: damien.saucez@inria.fr Saucez (Ed.) Expires April 18, 2013 [Page 12]