SDN: A NEW PARADIGM Kireeti Kompella CTO, JDI
AGENDA What is SDN? Definition and goals of SDN Analogy with Compute Virtualization Orchestration for Agile Provisioning Unified SDN What parts of the network does SDN touch? Conclusion 2 Copyright 2014 Juniper Networks, Inc. www.juniper.net
DEFINITION AND GOALS OF SDN SDN has many definitions and many goals Separation of control and data planes Interoperability, innovation, higher feature velocity Separation of control and data planes Programmability, more network control OpenFlow-based data plane Standardized, powerful Increasing use of CPUs (Intel x86 processors) General purpose, cheap, ubiquitous Next-gen provisioning: abstract service creation Lower OpEx, faster service delivery 3 Copyright 2014 Juniper Networks, Inc. www.juniper.net
SEPARATION OF CONTROL AND DATA PLANES CENTRALIZATION OF CONTROL PLANE à NOS Network Application Network Application Network Application Network State Transform via Network Apps Network Operating System Control Control Control Data Plane Data Plane Data Plane 4 Copyright 2014 Juniper Networks, Inc. www.juniper.net
IS A ROUTER JUST CONTROL AND DATA PLANE? Not standardized; not at all interoperable! This is where services live Interoperable (more so than not) This is also where agility is needed Mgmt Mgmt Mgmt Academic view of a router Control Data Plane Control Data Plane Control Data Plane Actuality of a router 5 Copyright 2014 Juniper Networks, Inc. www.juniper.net
RESTORING AGILITY: SEPARATE MANAGEMENT FROM REST OF ROUTER APIs are for defining and controlling the orchestration RESTful APIs Orchestration Layer The goal: service agility via orchestration: freedom from physics, process, bureaucracy Mgmt Mgmt Mgmt Control Control Control Data Plane Data Plane Data Plane 6 Copyright 2014 Juniper Networks, Inc. www.juniper.net
ORCHESTRATION = AGILE SERVICE PROVISIONING Just as in Compute Virtualization, so in networks: we need the ability to orchestrate and automate In this case, service creation and management This offers greater network control and speeds up service rollout Dual effect of an increase in revenues and lower OpEx It will also improve resource efficiency, leading to CapEx savings But how is this different from current Network Management Systems, or OSSs? 7 Copyright 2014 Juniper Networks, Inc. www.juniper.net
DISCUSSION: WHY PROGRAMMABILITY? Cool! New control plane! really? 8 Copyright 2014 Juniper Networks, Inc. www.juniper.net
ANALOGY: COMPUTE VIRTUALIZATION First came the technology (actually, quite long ago) Let s emulate virtual CPUs on a physical CPU Then came orchestration and automation Bring up manage, terminate, new VMs Finally, came the realization I m free from the constraints of physics! and the constraints of process, bureaucracy Instead of a few months to deploy an application agonize over how many servers are needed, purchase, rack and power them, connect them up it can take a few hours, or even a few minutes especially with the help of automation 9 Copyright 2014 Juniper Networks, Inc. www.juniper.net
PROGRAMMABILITY IS MOSTLY ABOUT AUTOMATION Automation allows repetitive tasks to be done easily, quickly and in an error-free manner Automation allows pre-programmed responses to events: faults, congestion, bandwidth surges, bandwidth requests, Automation allows humans to offload what they don t do well quick, accurate responses to anticipated events to get on with what humans do best: defining new services, defining policies, monitoring for unanticipated events and providing customized responses This means that the language in which this automation is expressed needs to be suited for the purpose 10 Copyright 2014 Juniper Networks, Inc. www.juniper.net
DISCUSSION: WHY PROGRAMMABILITY? Practical, efficient, gets the job done practical, efficient, sexy-ish 11 Copyright 2014 Juniper Networks, Inc. www.juniper.net
SERVICE ABSTRACTION definition is based on abstract information models These are high-level: device and OS and version independent They are standardized, but allow for provider-specific enhancements deployment is transformation of an abstract service definition to device-specific data models In this system, service deployment will be: Fungible I can use a given device for many services Flexible I can deploy a service at many devices (placement) Fast I can roll out a service quickly, redeploy quickly Provisioning Responsive the service adapts dynamically to changes Analytics and Automation 12 Copyright 2014 Juniper Networks, Inc. www.juniper.net
SDN AS A COMPILER SAY WHAT YOU WANT, NOT HOW TO DO IT service reqts High-level, declarative specification of service requirements configuration lives here S DB Process & compile SDN system Parse specification Process analytics Configuration is sent to chosen device A DB Network Analytics 1 2 3 4 5 6 13 Copyright 2014 Juniper Networks, Inc. www.juniper.net
UNIFIED SDN OSS/Orchestration model 1 model 2 model 3 Abstract SDN Transformation Engine model 1 model 2 model 3 level 1 2 3 4 n Access Edge/NFV/DC Core/Inter-DC 14 Copyright 2014 Juniper Networks, Inc. www.juniper.net
CUSTOMIZATION OF SERVICE MODELS AND DEVICE MODELS aka programmability Standard model for some service Template Customized service model for use case Customized Model Instance for some customer Customized Instance Standard device level model for service Base device model Vendor-customized device level model Customized device model model for service instance instance 15 Copyright 2014 Juniper Networks, Inc. www.juniper.net
SDN AS A COMPILER WITH PLUGINS TO HANDLE CUSTOMIZATIONS Instance Template Use case specific features Customized Template Customized Instance Plugin for specific use case Plugin for Base Compiler Plugin for specific resource optimization level model specific feature level model instance 16 Copyright 2014 Juniper Networks, Inc. www.juniper.net
AGENDA What is SDN? What parts of the network does SDN touch? Virtualized Data Centers SDN for NFV Core SDN Access/Edge SDN Conclusion 17 Copyright 2014 Juniper Networks, Inc. www.juniper.net
WHAT PARTS OF THE NETWORK DOES SDN TOUCH? Access SDN Edge SDN MOBILE ACCESS ACX1000 ACX4000/MX104 MX240/480 MX960 BUSINESS ACCESS ACX1000 CEN Access (CKT/ETH/MPLS) CEN aggregation (MPLS) CEN Core (MPLS) RESIDENTIAL ACCESS ACX1000 CELL-SITE / CPE ACX4000/MX104 CONVERGED ACCESS/ HUB-SITE ROUTER MX240/480 AGGREGATION ROUTER MX960 EDGE ROUTER Core Virtualized DC To other metros MUX/ DEMUX MUX/ DEMUX amp amp amp amp amp amp amp amp MUX/ DEMUX MUX/ DEMUX NFV WAN SDN: Centralized and Multi-Layer Traffic Engineering SERVICE CENTER 18 Copyright 2014 Juniper Networks, Inc. www.juniper.net
SDN FOR VIRTUALIZED DATA CENTRE Orchestration of Compute (Virtual Machines) in a Virtualized DC is mature One can, via the Contrail controller, define Virtual Networks (VNs), inter-vn policies and service chains, as high-level data models When a VM is placed on a server, the Network Orchestrator calls into to Contrail to connect the VM to its VN, and has all the policies it needs Virtual Machine Server Orchestrator Compute APIs Gateway Router JunosV Contrail Network APIs Physical Switches Internet VPN DCI WAN Storage APIs vrouter Nodes vsrx, F5 19 Copyright 2014 Juniper Networks, Inc. www.juniper.net
NETWORK FUNCTION VIRTUALIZATION NFV is the use of general purpose servers instead of specialized devices for network functions such as voice gateways, video encoders, IMS, MME, firewalls, DPI, etc. Virtualized Network Applications Virtualized Network Applications VMs VMs Host/ Cluster Manager KVM Hypervisor Host/ Cluster Manager KVM Hypervisor SERVICE CENTER Linux Linux 20 Copyright 2014 Juniper Networks, Inc. www.juniper.net
NFV ORCHESTRATION Abstract specification of services and service chain Servers Orchestrator Compute Models Network Models Specification of services includes details of instances and config; and details of the connecting service chain Instantiate services Gateway Router Create service chain DCI WAN VPN Internet 21 Copyright 2014 Juniper Networks, Inc. www.juniper.net
SDN FOR THE EDGE instances defined via service data models instance database Instance1 S DB Program devices Instance2 Process & compile Instance3 SDN system A DB Analytics database Real-time telemetry Self- Portal The value of automation! Agg2 CPE Access Network Edge Core 22 Copyright 2014 Juniper Networks, Inc. www.juniper.net
CORE (OR WAN) SDN: PATH OPTIMIZATION SDN in the core of the network focuses on optimized multi-layer Traffic Engineering based on centralized path computation There is a thoughtful separation of functions into those that are distributed and those that are centralized The goal is to have a higher-level view of TE, and to make this TE serve the needs of services 23 Copyright 2014 Juniper Networks, Inc. www.juniper.net
CORE SDN: MULTI-LAYER PATH COMPUTATION & BANDWIDTH CALENDARING Different application, different domain, but same SDN paradigm! PCEP Abstract, network-wide LSP reqts Management Node Distributed signaling POP POP POP POP POP Control Node Centralized Path Computation Distributed topology discovery BGP-TE POP POP POP Note: only a small subset of the paths (2 paths) are shown. Primary Backup thoughtful separation of functions into distributed and centralized 24 Copyright 2014 Juniper Networks, Inc. www.juniper.net
CONCLUSION: THE SDN PARADIGM The paradigm of SDN as a Network Operating System to form the basis for network programming is too low-level This attacks the problem of new control plane implementations Worthwhile problem for some, but not for all The paradigm of SDN as a Compiler for provisioning via abstract service models is a high-level, declarative approach This attacks the problem of service provisioning Real problem for most who provide and manage network services This paradigm is standards-based while allowing for providerspecific enhancements This paradigm applies to all parts of the network: DC, inter-dc, access, edge, NFV, core 25 Copyright 2014 Juniper Networks, Inc. www.juniper.net