The Impact of SDN On MPLS Networks Adrian Farrel Juniper Networks afarrel@juniper.net www.mpls2012.com Agenda • • • • • • • Some Definitions Needed (SDN) Why consider SDN? SDN as a toolkit Fine-grain programming An interface to routing and policy Enabling services A programmable MPLS network 2 What do we mean by SDN? • Software • It’s all software! • We are looking for automation • Tools or applications • Driven or Defined • Does it matter? • Networks • Micro-management of forwarding decisions • Control of end-to-end paths • Whole-sale operation of network • Is it all about the buzz? • Shiny-thing Desparately Needed • $ome Dollars Now 3 What will we do with SDN? • • • • • Make our networks better Provide cool services at lower prices Reduce OPEX and simplify network operations Enable better monitoring and diagnostics Make better use of deployed resources • Converged services are the future • Converged infrastructure is the future 4 The SDN Toolbox • To a network, all configuration tools look like hammers • One size does not fit all • SDN is about engineering the routers to allow better tools to be designed and applied 5 MPLS Data Plane Programmability • Label stacking, popping, and swapping • MPLS as a cross-connect technology (NHFLE) • Prefix-to-label mapping (LFIB) LFIB Prefix Out i/f, label IP Packet NHFLE In i/f, label Out i/f, label MPLS Packet NHFLE In i/f, label Pop MPLS Packet • Integration with underlying data plane • Encapsulation rules • Data link addresses 6 IP Packet Applications and Protocols for MPLS Data Plane Programmability • An application is software that runs remotely • That demands a protocol and a data encoding • Element Management • IETF : SNMP/MIBs, Netconf/YANG • Proprietary : CLI, GUI, XML • Legacy : TL1 … • Network Management • Coordinated connection set-up is just coordinated element management • Control plane / data plane separation • • • Use an existing configuration protocol IETF : GSMP (RFC 3292), ForCES (RFC 5810) ONF : OpenFlow 7 Control Plane / Data Plane Separation • • • • Support legacy / cheap devices Experiment with new routing protocols Integrate dynamic routing with static control Avoid “complexities” of routing protocols Control plane SDN Programming Language Data plane 8 Functional Control at a Higher Level • Operators want to build and deploy services • • • • • • • “Make a pseudowire for me” “Optimize my traffic loading” “Provision a layer 3 VPN” “Show me how my network is being used” “Configure my data center” “Manage security and policy” “Provide service callendaring” • Needs a higher-level interaction with the network • Demands more sophisticated control of routers • Must integrate with standard routing features 9 Leveraging Existing Tools • • • • New services and features for rapid deployment There are plenty of existing tools Leverage implemented and deployed protocols We can put them together to enable high function SDN in MPLS networks • May need some extensions • Avoid long development cycles 10 BGP-LS to Extract Topology Information • Information about the network • • • • • Nodes and links Link state Up-to-date TE capabilities Delay and other quality information Status of existing LSPs / tunnels • Used for network monitoring, analysis, and planning • Critical input to path computation (e.g., via PCE) • Fundamental component unspecified in the PCE architecture • BGP-LS is a set of simple extensions to BGP • Client is any node listening to the IGP • For example an ASBR or a Route Reflector • Server can not be a very light-weight BGP implementation • Reduces dependency on IGP sniffing 11 Stateful PCE for Control of Services • Early work on PCE was stateless • PCE knows state of network • PCE does not recall anything about previous computations • PCE does not know about existing provisioned services • Except as described on new computation requests • Stateful PCE was always in the architecture • Retain information about provisioned LSPs • New extensions to PCEP • Allow explicit activation of LSPs from the PCE • Receive information from network about LSPs • Provides key components for bandwidth callendaring 12 Integrating the Components • • • • • IGP enhanced for TE and link quality BGP-LS reports to PCE PCE requests LSPs Normal LSP signaling LSP status reports PCE 13 Making New Tools • Can’t do everything with what have already • Interface to the Routing System (IRS) • A programmatic interface to routers Application Application Application Server IRS Client IRS Client IRS Protocol & Data Encoding Router OAM, Events and Measurement Policy DB Data Plane IRS Agent RIBs and RIB Manager Topology DB Routing and Signaling Protocols FIB 14 14 Enabling Services • Service enablement and turn-up is complex • Existing tools help with planning • Commissioning through scripts or work-plans • SDN can be a set of tools to enable services • • • • L3VPN delivery Data center interconnect Bandwidth callendaring Mult-layer connectivity and virtual links 15 Service Example : Multi-layer • SDN can coordinate multiple network layers • May both be MPLS networks • Involves many SDN components Traffic demand Service request IRS BGP-LS TEDB IGP-TE PCE PCEP IRS Policy IRS Virtual Link RSVP-TE VNTM OpenFlow & IRS PCEP IRS GMPLS PCE TEDB 16 IGP-TE Service Example : L3VPN with Callendaring • • Which PEs to use? How to connect PEs? • • • • What load? When? What redundancy? QoS? Security? How to connect to the Internet? Planned support for high bandwidth services DB Replication Content Streaming Data Transfer 17 Putting the Tools into the Box • SDN will possibly remain buzz and hype • Or maybe it will evolve into bickering between proponents of different solutions • Or it could become a comprehensive set of tools • • • • Configuration tools RIB and policy control Topology and LSP management Service enablement • Potential to enable a rich set of functions in future MPLS networks 18 SDN - Pandora’s Toolbox? A mess of overlapping tools and protocols with too many features and functions? Or a cornucopia of riches? 19 Questions? afarrel@juniper.net 20