Virtual Private Networks: Progress and Challenges Panel Session Copyright © 2000, Juniper Networks, Inc. Panel Objectives • Introduce Virtual Private Network concepts and technologies • Describe some potential service provider VPN offerings • List challenges faced by service providers in offering VPN services • List and describe some of the proposals for addressing VPN challenges Copyright © 2000, Juniper Networks, Inc. Slide 2 Panel Participants • Paul Ferguson – Cisco Systems • David O’Leary – Juniper Networks • Keerti Melkote – Nortel Networks • NANOG audience (Question and Answer) Copyright © 2000, Juniper Networks, Inc. Slide 3 Virtual Private Networks: Progress and Challenges David O’Leary Director, Consulting Engineering Copyright © 2000, Juniper Networks, Inc. What is a VPN? • Virtual • • • Emulation of a private network facilities over a shared network infrastructure Private • Minimally: no mixing with traffic outside the VPN, and support for private address space(s) • Possibly encryption and protected traffic class Network – two or more users or sites Copyright © 2000, Juniper Networks, Inc. Slide 5 How Virtual is Virtual? • The only true non-virtual private network are customer-owned physical plant, like copper and fiber, transport and switching equipment • Leasing TDM circuits from a carrier means that the customer gets a “virtual” slice of the carrier’s transmission network • Leasing some kind of layer 2 circuits (ATM, Frame Relay) from a carrier means that the customer gets a “virtual” slice of the carrier’s layer 2 network • Statistical multiplexing here means that it’s cheaper for both the provider and (in theory) the customer Copyright © 2000, Juniper Networks, Inc. Slide 6 Focus on “IP VPNs” • VPNs over an IP backbone that supports multiple services (e.g., public Internet, VoIP) • Exploit economies of scale through use of common backbone facilities • Reduce inefficiencies of separate networks • Shared local loops for internal corporate network and Internet access • Service providers add value by allowing customers (enterprises networks) to “outsource” their routing (complexity) to the carrier Copyright © 2000, Juniper Networks, Inc. Slide 7 Four models of VPNs •Remote User access •CPE Based •MPLS-based Layer2 •Provider-Based Layer 3 Copyright © 2000, Juniper Networks, Inc. Slide 8 Remote User Access • Variety of protocols developed in mid-90’s to tunnel remote user traffic to a fixed site on the IP network • ATMP, PPTP, ATMP • Functions consolidated in IETF L2TP protocol • Documented in RFC 2661, with various drafts for extensions • Dynamic, authenticated tunnels • Deployments are becoming quite common Copyright © 2000, Juniper Networks, Inc. Slide 9 CPE Based VPNs • Tunnels configured between CPE devices • Options are GRE, IPSEC, IP-in-IP, PPTP, L2TP • Topology of the VPN is configured into the CPE devices • The provider does not have to do anything to their network (they may not even know it is happening) • VPN traffic may be marked and prioritized • The service provider may bundle and manage the CPE devices used to create the VPN service • Provides “value add” beyond best effort Internet connectivity • This model is already being aggressively deployed by providers around the world Copyright © 2000, Juniper Networks, Inc. Slide 10 CPE Based VPNs Tradeoffs • • Provider network configuration • Potentially no more than required for Internet access • CoS or managed service adds provider complexity Customer (CPE) configuration • • Every tunnel is a separate virtual interface that must be configured, including for routing Scalability • Provider network – excellent • Customer – depends on number of tunnels and routing topology/complexity Copyright © 2000, Juniper Networks, Inc. Slide 11 Layer-2 VPNs • • A subscriber leases VCs between the sites that need to be connected • Topologies are hub and spoke, full or partial mesh • The subscriber and provider think of these VCs as “dumb” pipes not at all involved in Layer 3 issues such as routing, packet filtering, etc. • Subscriber outsources the Layer 3 management to the provider in a “Managed Router” service Mature support for commitments about service (e.g., bandwidth, availability, etc.) • At least in the business/SLA sense • Virtual Circuit model eases capacity planning Copyright © 2000, Juniper Networks, Inc. Slide 12 Layer-2 VPN Tradeoffs • Provider configuration • • • • A lot in theory, but it’s mostly automated in practice Subscriber configuration • Every VC is a separate virtual interface that must be configured, including for routing • Large, complex topologies yield complex configurations Scalability • Provider -- number of VCs and stability of core • Subscriber -- number of interfaces and routing Other • Leverages existing investment Copyright © 2000, Juniper Networks, Inc. Slide 13 MPLS-Based Layer-2 VPNs • Looks identical to “traditional layer 2 VPNs” from the subscriber perspective • Provider carries the layer 2 circuits over an IP/MPLS backbone • Provides the ability for multiple services (public IP, private IP, VoIP, etc) over a single access circuit Copyright © 2000, Juniper Networks, Inc. Slide 14 MPLS-Based Layer-2 VPNs Internet Traffic: ATM VC1 terminated, IP packets delivered to provider 2 Provider 2 Provider 1 Subscriber A ATM Access ATM Access Subscriber A VPN Traffic: ATM VC2 mapped to MPLS LSP “tunnel” Termination of ATM PVCs (layer 3 lookup) and support Layer-2 passthrough on the same port. Copyright © 2000, Juniper Networks, Inc. Slide 15 MPLS-Based Layer-2 VPN Tradeoffs • Provider configuration • • • Manual configuration of ingress and egress boxes (could be partially automated) Subscriber configuration • Every VC is a separate virtual interface that must be configured, including for routing • Retains existing CPE and subscriber model Scalability • Provider -- number of LSPs, stability of core • Subscriber -- number of VCs and routing Copyright © 2000, Juniper Networks, Inc. Slide 16 Provider-Based Layer-3 VPNs • Concepts outlined in RFC 2547 and RFC 2764 • Subscriber treats access link as combined internet/VPN link • Except for multiply-connected sites, needs minimal configuration (i.e., Default) • Provider’s edge router supports instances of routing protocols and multiple forwarding tables • • If a destination isn’t in a VPN-specific forwarding table then use Internet table VPN site membership and VPN-specific routing information carried in BGP • Supports overlapping private address space • Topology is conceptually always a full mesh between PE’s Copyright © 2000, Juniper Networks, Inc. Slide 17 Provider-based Layer-3 VPN tradeoffs • Provider configuration • Minimal static configuration in basic scenarios • More configuration needed for complex topologies • Subscriber configuration • Basic scenario is straightforward • More complex config needed for security, multihoming, participation in multiple VPNs, etc. • Scalability • Provider • I-BGP possibly sees 100,000s of routes, (in)stability of multiple dynamic routing domains, multiple forwarding tables • Subscriber – allows shared local loop and CPE device for multiple services (intranet, Internet, voice, etc.) Copyright © 2000, Juniper Networks, Inc. Slide 18 Where Does That Leave Us? • Customer-centric tunneling are easier for the provider • COS, traffic engineering, SLA’s, etc. are value-adds • For a provider implementing VPNs, scalability and value-add are fundamentally at odds • MPLS-based layer-2 VPNs offer the benefits of an integrated/multi-service network • But the enterprise network has to handle the routing • Layer-3 VPNs allow the enterprise network to be almost unconcerned with routing • Doesn’t support as many VPNs, sites per VPN, routes per VPN or customers per access box than MPLS-based layer 2 VPNs • Risk of the stability of the multi-service core • So what is the answer? Copyright © 2000, Juniper Networks, Inc. Slide 19 Where Does That Leave Us? (cont.) • The answer is: it depends • Some subscribers may require the ability to outsource routing • • Layer 3 VPNs or • MPLS-based layer 2 VPNs where “managed services” are provided and a device at the customer’s premises converts from pure IP to some number of layer 2 virtual circuits Some subscribers may be capable of doing their own routing • • MPLS-based layer 2 VPNs offer scalability for broad deployment as well as stability with less coupling Range of hybrid solutions is probably likely • Remote access user VPNs • CPE-based VPNs • MPLS-based layer 2 VPNs • Layer 3 VPNs Copyright © 2000, Juniper Networks, Inc. Slide 20 Summary • There is no one-size-fits-all solution for VPNs • Tradeoffs between value add and scalability/stability • Tradeoffs between customer requirements • Desire to outsource routing, security, etc. • Size and complexity of network Copyright © 2000, Juniper Networks, Inc. Slide 21