NEURAL NETWORKS Biological analogy Introduction to Artificial Neural Networks Typical architectures.
New Control Architectures for E2E networks
description
Transcript of New Control Architectures for E2E networks
New Control Architectures for E2E networks
Juan Pedro Fernandez-Palacios, Telefonica I+D ([email protected])
April 2013
Core network operation is not adapted to flexible networking Multiple manual configuration actions are needed in core network nodes Network solutions from different vendors typically use particularized Network
Management System (NMS) implementations Very long service provisioning times
Internet Voice CDN Cloud BusinessService ManagementSystems
Network ProvisioningSystems
Metro NMS
NMS Vendor A
IP Core NMS
Optical Transport NMS
Umbrella Provisioning System
Complex and long workflows for network provisioning over different segments (metro, IP core, Optical transport) requiring multiple configurations over different NMS
NMS Vendor B
NMS Vendor C
NMS Vendor D
NMS Vendor E
NMS Vendor C
NMS Vendor A
NMS Vendor B
Metro Node
Vendor A
Metro Node
Vendor B
IP Node
Vendor C
IPNode
Vendor D
IPNode
Vendor E
Optical Node
Vendor A
Optical Node
Vendor B
OpticalNode
Vendor CCore Network Nodes
CURRENT APPROACH FOR NETWORK PROVISIONING
Traditional core network operation is very complex and expensive
Control plane and SDN pave the path towards a unified network provisioning architecture
Key building block of such unified network provisioning architecture are: Network configuration interface: Multivendor edge nodes configuration (e.g
OLT and BRAS, IP core routers, etc) by standard interfaces (e.g OpenFlow) IT and network SDN orchestration: Coordinated network and datacenter
resources control according to service requirements (e.g orchestrated Virtual Machine transfer among datacenters)
Network-Service API: Application level API hiding details of the network
Internet Voice CDN Cloud Business
Multiservice network provisioning system(SDN Orchestrator)
Standard signaling mechanisms running over network nodes enabling flexible networking and automated network
provisioning over different network segments (metro, core IP, optical transport) including multiple vendors
Metro Node
Vendor A
Metro Node
Vendor B
IP Node
Vendor C
IPNode
Vendor D
IPNode
Vendor E
Optical Node
Vendor A
Optical Node
Vendor B
OpticalNode
Vendor C
Service ManagementSystemsNetwork Provisioning
Core Network Nodes
Network-Service API
Network configuration
interface
Unified network provisioning architecture
Infrastructure Layer (e.g DataCenter)
SDNController
ALTO SDN orchestrator
Application Layer
OAM Handler
TED VNTM PCE
Provisioning Manager
OPENFLOW
SDN CONTROLLER
API
OpenFlow is based on the concept of actions that are applied to each packet of a given flow (Ethernet-level addresses, VLAN tags, IP
addresses, MPLS labels or transport-level ports).
The actions taken by SDN the controller comprise: inserting and removing tags (layer 2), performing routing (layer 3) and also providing
differentiated treatment to packets (QoS)
Basic SDN Approach for OpenFlow Domains
82nd IETF, Taipei
1) Discovery of network resources2) Routing, path computation3) Automated network orchestration in response to changing network
conditions and service requirements 4) Network resources abstraction to application layer5) QoS control and performance monitoring 6) Multilayer interworking7) Multidomain/multivendor network resources provisioning through
different control domains (e.g OpenFlow DataCenter, OpenFlow MAN, GMPLS optical transport…)
E2E networks might be pure OpenFlow based one day, but the migration process will take
some time
Main actions to be taken by the SDN controller in E2E networks
OpenFLowData
Center
SDNController4-ALTO3-SDN orchestrator
Applications (Internet, CDN, cloud…)
5-OAM Handler
1- TED 6-VNTM 2-PCE
7-Provisioning Manager
OpenFlowMAN
Domain
IP/MPLScore
OpenFlowOptical Domain
OPENFLOW OPENFLO
W
GMPLSOptical Domains
NETCONF
MPLSMAN
PCEP OPENFLOW CLI
Most of these building blocks are still on
definition and standardization
process
SDN controller based on standard building blocks
Physical Network
NETWORK OPERATING SYSTEM
CSO Multilayer Orchestrator
Link Provisioni
ng
CDN and nionetwork optimizat
APIAPI API
Provisioning Manager
Cloud Services Live OTT Internet…
Orchestration mechanisms (*)
Network APIs
…
NetConf OpenFlow
PCEP UNI
Inside SDN Orchestrator
Access Network Metro Area Network Core Network
Optical Transport
Multidomain L2 service provisioning
Data Center Network
CPEVirtual
Machine (e.g BRAS)
Technical challenges: Horizontal Orchestration. Automated L2 service provisioning
through different packet switching domains (metro, core, datacenter).
Vertical Orchestration. This orchestration enables adaptive network resources allocation in IP and optical layers according to the traffic pattern to efficiently use network resources
Multilayer orchestration
SDN controller
E2E SDN
control
Access Network Metro Area Network Core Network Data Center Network
CPEVirtual
Machine (e.g BRAS)
SDN controller
OpenFlowCLI CLI
Multidomain pseudowire over seamless MPLS
Intra datacenter connection
Multidomain L2 service provisioning (short term)
Access Network Metro Area Network Core Network Data Center Network
CPEVirtual
Machine (e.g BRAS)
SDN controller
OpenFlowOpenFlow OpenFlow
Multidomain pseudowire over seamless MPLS
Intra datacenter connection
Multidomain L2 service provisioning (Medium term)
For this scenario, OF is used to trigger control plane. This means that edge nodes have to decode OF and translate into CP messages.
For the case of creating a Pseudo-Wire following parameters are required:• Pseudowire Label
• MPLS Label
• Service VLAN (VLANs)
• Output port
11
CP node
OF and CP node enable node
OF Request
1
3OF Information Updated
2
4
Multidomain L2 service provisioning (Medium term)
Access Network Metro Area Network Core Network Data Center Network
CPEVirtual
Machine (e.g BRAS)
SDN controller
Intra datacenter connection
OpenFlow
Common Interface
OpenFlow
SDN controllerSDN controllerSDN controller
Options: Hierarchical Approach. There is a controller which has a
global view so it can orchestrate the configuration in each domain.
Peer Relationship. Each controller can request for information or connections to other peers.
Connection to datacenter
Multidomain L2 service provisioning (Long term)
Access R1
Access Region 2
Transit R1
Transit R2
TransitBackup R3
Interconnection
Transit R3
Access R3
• Increased survivability• Extended reparation processes• Capex Savings (best effort
traffic only)
Load balancing between IP and optical networksMulti-layer restoration
Vertical Orchestration
OpenFLowOPS Data
Center
SDNController4-ALTO3-SDN orchestrator
Applications (Internet, CDN, cloud…)
5-OAM Handler
1- TED 6-VNTM 2-PCE
7-Provisioning Manager
OpenFlowMetro-Core
Node (L3/L2/L1)
IP/MPLScore
OpenFlowWSON
network
OPENFLOW, GMPLS
OPENFLOW
GMPLSFlexiGrid
NETCONFPCEP, GMPLS OPENFLOW OFELIAIDEALISTIDEALISTDISCUSSTRAUSS
IDEALIST: IP and Flexgrid configuration
IDEALISTIDEALISTIDEALIST
IDEALIST: Multilayer IP over FlexiGrid Orchestration
STRAUSS: VM transfer orchestration
EU projects situation in this picture
E2E SDN control (KDDI, NTT, NEC…)Network Operating System Multilayer and multidomain orchestration mechanismsNetwork Functions Virtualisation
Optical data plane (NTT, Fujitsu, Osaka University, NEC…)Subwavelength, Flexgrid, Optical OFDMSliceable and Programmable Transponders
“sliceable” BVT. Figure from NTT.
Joint EU-Japan standardization contributions (IETF, ONF, NFV, ITU…)
List of potential topics for future collaboration EU-Japan