SDN/NFV: Enabling Programmability in Networks Is this end game for hardware?

SDN/NFV: Enabling Programmability in Networks Is this end game for hardware? Dimitra Simeonidou High Performance Networks, University of Bristol, UK 1

Key Enablers for Network Programmability SDN and NVF enable programmability using control/management/service software tools and assuming hardware pre-defined functionality. What needs to happen at the hardware level? Unlocking flexibility and programmability at software and hardware layers 20 M UK research portfolio (Engineering and Physical Science Research Council: www.epsrc.ac.uk) Toward Ultimate Convergence of All Networks (TOUCAN) Transforming the Internet Infrastructure: The Photonics HyperHighway Introducing Insight into the Abstraction of Optical Network Infrastructures (INSIGHT) Knowledge Centric Networking (KCN) Synthetic On-Chip and Off-Chip Optical Network Design (SONATAS) Universities: Bristol, Edinburgh, Herriot Watt, Lancaster, Southampton, Surrey, UCL Industry:BBC, Broadcom, BT, NEC, Polatis, Samsung Portfolio of 9 EU (FP7, H2020 projects on DC architectures, 5GPPP) 75M Smart City project: Bristol Is Open

EPSRC Programme Grant: Towards Ultimate Convergence of All Network: (TOUCAN) Non-academic Partners: BCC, Broadcom, BT, Janet, NEC, Innovate UK, Plextek, Samsung Advisory Board: John Manley (Chair), P. Demeester (iminds), S. Figuerola (i2cat), Y. Wolfsthal (IBM, Israel), A. Gladisch (Deutsche Telekom), J. Elmirghani (INTERNET, PI), C. Chambers (BBC), EPSRC Funding: 6M from the UK Research Council (EPSRC) + 6M from partners and Institutions Duration 5 yrs from 11 th of August 2014

Key Message A1 TOUCAN will develop a NETWORK TECHNOLOGY USBwhich will enable any network device to be plugable, discoverable, describable and interoperable in the network

Slide 4 A1 Add outdoor LiFi Anna, 09/09/2014

TOUCAN aims to achieve ultimate network convergence enabled by a radically new technology agnostic architecture targeting a wide range of applications and end users Facilitate seamless interconnection of network technology domains, networked devices and data sets New foundations for end-to-end convergence High flexibility in data throughput, high adaptability, resource efficiency Address Quality of Service (QoS) and Quality of Experience (QoE) requirements Enable application-level infrastructure programmability

H2020 5GPPP5: G-XHaul Infrastructure Flexiblenetwork architecture able to transport future RAN solutions distributed and dense small cells networks, centralized Cloud-RAN deployments, hybrid etc. Fully programmable Time Shared Optical Network (TSON) is adopted to provide connectivity between: macro-sites or central offices and the regional DCs (BBUs) in case of Cloud-RAN core network gateways in case of a distributed RAN architecture

The Solution How to solve the problem? Manage a diverse Network-of-Networks IT, Telecoms, Wireless, Sensors, Actuators, IoT, etc. Multi-Technology Support Virtual Network A Virtual Network B Network Operating System (NetOS ) Computers Software Defined Networking Abstractions of the underlying hardware Logically centralized control Eliminate Vendor Lock-ins white box solutions? Data centres & Cloud Network Virtualization Slice network to share resources among many users Efficient utilization & monetization Multi-Tenant Solution Sensors, M2M, IoT Telecoms INFRASTRUCTURE

An SDN Enabled Architecture: for Network Control and Virtualization Network Virtualisation Node Slicing Bandwidth Slicing Network Abstraction VSwitch Uniform switching VSwitch VSwitch VSwitch Switch as a VSwitch Information Model Device: Features Actions Peering TOR Space Switch Advanced NIC Fast switch WSS OF Agent

Support for any arbitrary network elements Supporting Technology and Architecture Evolution Support for fixed devices but also flexible devices with switching capability Advanced NIC TOR Space Switch Fast switch WSS

Support for any arbitrary physical connectivity Supporting Technology and Architecture Evolution Flexible to change connectivity between devices during architecture evolutions Advanced NIC TOR Space Switch Fast switch WSS

An SDN Enabled Architecture: for Multiple Network Technologies OF Ex. OF Agent Extended OpenFlow Channel Capability Abstraction (Resource Model) Technology specific interface TL1 Raw Socket Raw Socket Ethernet Frame Space Switch OPS Switch Flex-grid WSS FPGAbased NIC

Abstraction and Virtualisation

Physical Layer Impairment Assessment Model Shuping Peng, Reza Nejabati, Dimitra Simeonidou, Impairment-Aware Optical Network Virtualization in Single-Line-Rate and Mixed-Line-Rate WDM Networks, IEEE/OSA Journal of Optical Communications and Networking (JOCN), 5(4), pp. 283-293, April 2013.

Network Functions Virtualisation: NFV Network Functions Virtualization is about implementing network functions in software -that today run on proprietary hardware - leveraging (high volume) standard servers and IT virtualization Supports multi-versioning and multi-tenancy of network functions, which allows use of a single physical platform for different applications, users and tenants Facilitates innovationtowards new network functions and services that are only practical in a pure software network environment Applicable to any data plane packet processing and control plane functions, in fixed or mobile networks Source: Adapted from D. Lopez Telefonica I+D, NFV

Classical Network Appliance Approach Network Functions Virtualisation: NFV Network Virtualisation Approach Independent Software Vendors Message Router CDN Session Border Controller WAN Acceleration Orchestrated, automatic & remote install. DPI Firewall Carrier Grade NAT Tester/QoE monitor Standard High Volume s Standard High Volume Storage SGSN/GGSN PE Router Optical Switch Radio Access Network Nodes White Box [Optical or Layer 2 or Wireless Access]

NFV & SDN & Virtualization Globecom 2013 Applica on Layer Virtual PCE Extended OF Controller Extended OF Controller vnet1 Virtualiza on Layer vnetn Op cal FlowVisor PCE Policy Virtual PCE Topology QoT N F V contr oller Network Func ons Slicer Topology Controller Classifie r Node Constraints Flow Message Ctrl Op cal Topology Planner & QoT es mator Extended OF Protocol Based on [OF 1.0 v0.3] 1 x Calient DiamondWave Fiber switch OpenSource GMPLS controlled Virtualization to 4 switches iplc-roadm OpenFlow Switch WSS-ROADM Directionless 4ch add/drop 2 x 1GE (SFP) 2x10GE (XFP) DWDM WSS-ROADM Fixed 4ch add/drop 2 x 10GE (SFP+ DWDM) University of Bristol, UK Flexible wdm devices OpenFlow Switch

SDN Allows decoupling of DP and CP. Network OS with apps Limits: Assumes Hardware Infrastructure is nonprogrammable Function Rigid SDN, NFV and Optical Layer NFV Deploy virtual function on commodity servers Obvious limitations to type of virtual functions possible? Network (Opto- Electronic) Infrastructure Currently non-programmable and function rigid

SDN, NFV and Network Function Programmable (NFP) Infrastructure SDN Allows decoupling of DP and CP. Network OS with apps Limits: Assumes Hardware Infrastructure is nonprogrammable Function Rigid Network (Opto- Electronic) Infrastructure Why not network function programmable (NFP) hardware infrastructure? NFV Deploy virtual function on commodity servers Obvious limitations to type of virtual functions possible?

Optical Backplane Physical layer Considerations Architecture-on-Demand Node 1 Fibre inputs/outp uts ToR 2 N ToR ToR Splitter 1 2 P 1 2 P AWG 1 2 P AWG 1 R 1 T OPS switch Network Fabric becomes programmable Heterogeneous pluggable entities Different NIC technologies Different TOR technologies Different Switching technologies Different transponder Different server racks Multi-dimension network : Space, Frequency, Time

One Hardware any Function Optical backplane(s) large scale port switch e.g. MEMS / Beam steering, Several pluggable photonic sub-systems and components L2 Swtich Freq Select MUX/DEMUX Fast (nsec) Switching EDFA Optical Backplane Fibers In/Out (Single- Core/Multi- Core/Vortex/ ) Fibre/Core switching Multicast/ Broadcast Fibers In/Out

One Hardware any Function FPGA-based Electronic Backplane Network function blocks Node Function blocks Physical technology drivers and transmission blocks FPGA-based Electronic Backplane Resource Controller Switching functions Ethernet/ TDM Switchover Ethernet Switch Network function blocks Control comm interface Node Function blocks FPGA Nodes Synchronization System operations Aggregation TDM Frames/slots Elastic TDM Switch QoT Overhead Switch control Control Physical technology drivers And Transmission Multicasting Traffic Optical Backplane Amplifi cation Multica sting MUX/D EMUX Fast Switch

Experimental Demonstrations

NFP ROADM Supporting TransportNetworking Scalable to SDM capacity levels thanks to efficient support for coarse granularities Flexibility to support fine switching granularities in space/frequency/time Support for additional functionality, e.g. defragmentation, regeneration, etc. More efficient use of functional modules: only used where required. λ NFP Node λ λ Fiber/Core λ λ TDM channel SDM DEMUX SDM MUX λ

Fully Programmable Optical DC Fabric (http://www.ict-lightness.eu) Agent Ethernet Frame ToR1 (FPGA) OVS OVS OVS Rack1 Agent 10G SFP+ Agent Raw socket OPS Back Plane (192 192 Polatis) ToR2 (FPGA) OVS OVS OVS Rack2 Agent ToR3 (FPGA) OVS OVS OVS Rack3 extended OF msg. TL1 Agent Agent SDN Controller Plugin OF lib with proposed extension OF lib with proposed extension Service Abstraction Layer Forwarding Rules Manager Topology Manager Topology Manager Extended modules REST Application

ECOC 2014 PD (University of Bristol and ORC Southampton) PLZT-based TDM SDM/WDM Converter Inter-DCN WSS Programmable FPGA-based ToR 40Gbaud QPSK/16QAM Tx Slotted-TDM/Ethernet programmable Tx Reconfigurable Optical Cross Connections MEF 1 MEF n Inter-cluster Top of Cluster Network Function Programmable switch 3 spools of 3-element fibers (MEFs) Optical 10GbE 25/19 ToR 1 10GE link Ethernet Traffic Analyzer ToR 2 Cluster Node 1 ToR n 4 4 Fast switch (10ns) Synchronized with TDM-enabled ToRs

Bristol Is Open Programmable City Experimental testbed where the University s opensource Network Operating System (NetOS ) controls a hardware programmable city infrastructure

Bristol Is Open (BIO) We are building a research network integrating optical, wireless, IoTand computing to provide a unique open and programmable experimental platform in the centre of Bristol BIO will offer the test-bed as utility for experimentation: City Experimentation as a Service BIO will support City-driven research and innovation

The Bristol Is Open Network Optical Network Wireless Network IoT Network Cloud Infrastructure 144-fiber core network connecting 4 nodes 1Gbps access network Wi-Fi, LTE, LTE-A, 60Ghz Fiber-connected lamppost Clusters & canopy of sensors HPC and commodity compute and storage A diverse network with different technologies 28

Smart City OS (NetOS ) end-user services core services experimental services App 1 App 2 App 3 App N NetOSMgmt Platforms management Exp1 Service Development Kit Enhanced SDK Data Presentation Layer Data Pre-Processing Services service 1 service N Platform Virtualization/Slicing Information Models Data Collection Engine sync async OpenStack OpenStack, etc. City Network(s) SDN Controller Gateway /Driver Gateway /Driver Gateway /Driver Gateway /Driver Gateway /Driver Gateway /Driver parking traffic gates smart grid public transport smartphones cars HPC Optical CCTV public buildings weather sensors... home meters sensors Wireless city resources citizen resources aux resources

Core Network Node User Boxes (NetFPGAs) OpenStack Managed DC in a Box SDN enabled L2 Switch 10 G/40G/100G SDN enabled L1/L0 Optical Switch 100G

Work In Progress Unlocking programmability through function coordination and co-design at software and hardware layers Telecoms Data Centres Smart Cities Sector specific benefits and boundaries Towards white box deployment, when & where?

Thank you