The Extension of Software-Defined Networking to Transport Networks Jim Theodoras Dec 2013
Introduction to SDN 2
SDN Oversimplified We like to simplify network diagrams. Today, networks have distributed control. SDN centralizes the control. Control Control Data Data Control Data Data Control Data Data Control Data Data The age-old argument of distributed vs centralized control 3
How Routers Talk Routers control messages are serial. Routers can only see nearest neighbors. Both allowed rapid expansion of networks. 4
SDN in a Single Sentence The hope/expectation of better efficiency/utilization through identifying flows of packets and making smarter holistic decisions A good analogy would be the Air Traffic Control system * http://mattsflight.blogspot.com * http://virtualskies.arc.nasa.gov/atm/6. html * flightradar24.co m 5
Any Cons to Centralized Control? 6
SDN is not really new AT&T s SS7 Signaling Network for 800 Phone Calls 7
2004 SIGCOMM'04 Workshops, Aug/Sept 2004, Portland, Oregon, USA 8
2007 SIGCOMM 2007: Centralized, reactive, per-flow control 9
2009 10
SDN Inside and Outside Data Center 11
Inside vs Outside the Data Center Inside Google s Iowa Data Center Outside Google s NC Data Center 12
Inside the Data Center Inside Google s Iowa Data Center Virtualizing Switching Storage and Compute are virtualized. Switching is only thing remaining. Common Management Different Vendors Different Functions One Control Commoditized Routers Open hardware platform. Dumb hardware does as told. 13
End Game 14
Outside the Data Center Outside Google s NC Data Center Smarter Transport Switching More dynamic use Less overprovisioning Common Management Among Transport Vendors Among Switch Vendors Among all equipment in network Virtual Sandbox for 3 rd Party Apps Open Layer 1 Transport to innovation. Ability to host simultaneous sessions. 15
Benefits of Data Center Interaction Decades of work have yielded today s agile core networks Unfortunately, the information to make intelligent decision resides at higher layers Problem made worse by today s flow dominated traffic Hybrid EDFA/RAMAN Amp Gridless ROADM Intelligent MUX Router Coherent Receiver Agile Core Network Router 16
Extending SDN to Transport 17
How to Extend SDN to Transport? Direct control model SDN Controller 18
How to Extend SDN to Transport? In-Direct control model SDN Controller 19
Breaking a Huge Task into Layers Transport Layer Switching Layer Orchestration Layer 20
DWDM Shelf Breaking a Huge Task into Layers Transport Layer Switching Layer PTX-LR2 Orchestration Layer 100G OTN 10GE DWDM Shelf PTX-LR1 PTX3000 10GE FSP3000 DWDM Shelf 100G OTN 10GE 100GE 21
Breaking a Huge Task into Layers Transport Layer Switching Layer Orchestration Layer 22
Where Transport Fits in SDN Model Hypervisors allowed Storage and Compute resources to be managed together. Hypervisor CPU Storage Compute 23
Where Transport Fits in SDN Model SDN allowed Networking to join in the fun through an Orchestration layer. Orchestration Hypervisor SDN Storage CPU Compute Network 24
Where Transport Fits in SDN Model Applications can be written on top of Orchestration layer. User Interfaces 3 rd Party Apps Orchestration Hypervisor SDN Storage CPU Compute Network 25
Where Transport Fits in SDN Model Transport is added by extending the SDN controller function. User Interfaces 3 rd Party Apps Orchestration Hypervisor SDN Controller Storage CPU Compute Network 26
Where Transport Fits in SDN Model A Network Hypervisor virtualizes the Transport network and presents an abstracted representation to the SDN controller. User Interfaces 3 rd Party Apps Orchestration Hypervisor SDN Controller Network Hypervisor Storage CPU Compute Network Transport 27
Where Transport Fits in SDN Model Now that all network layers are included in the control model, end-to-end, new applications are possible. User Interfaces 3 rd Party Apps Future Awesomeness! Orchestration Hypervisor SDN Controller Network Hypervisor Storage CPU Compute Network Transport 28
Where Transport Fits in SDN Model The Network Hypervisor has a big job in managing all Transport. User Interfaces 3 rd Party Apps Future Awesomeness! Orchestration Optical Network Hypervisor Hypervisor SDN Controller Optical Network Controller Management Control Database Storage CPU Compute Network Network Hypervisor Fault & Alarms Configuration Accounting Performance Topology Disc Path Compute Provisioning Resource Mgr Flow DB Topology DB Resource DB Security Policy Mgr Policy DB Transport 29
Transport in OpenStack Public Applications RESTful API Private Applications RESTful API Application Layer Openstack Orchestration cinder nova neutron Neutron plug-in Orchestration Layer REST VirtualNetwork Filter LLDP SDN Controller OPENFLOW OPENFLOW Network Hypervisor Control Layer 10010 vnic Resource Layer Storage Pool Compute Pool Network Pool Transport Pool 30
Real-World Applications 31
Proactive Bandwidth Calendaring Provisioning for peak traffic is losing battle, and only getting worse. Answer is dynamic network infrastructure. Site A Site B Daytime Configuration All Offices/Sites working 1x 2x 10G 2x 1x 10G 10G 2x 1x 10G Site C Nighttime Configuration Backup between A/B Double the bandwidth Other Configurations Site B to C Site C to A 32
Reactive Bandwidth Management Telefonica I&D IP Offloading Manager When traffic exceeds a threshold, an additional link is provisioned New circuit provisioning time Configured Threshold Rollback Threshold Circuit deprovisioning time 49 second bring-up 4 second teardown Fully automatic 33
Juniper, ADVA Dynamic Network Testbed Juniper MX-80 routers, ADVA FSP 3000 WDM Fully automated dynamic bandwidth-on-demand driven by router load variations Advanced Network Architectures Lab, Technical University of Catalonia, Juniper, ADVA Optical Networking 34 Published : OFC/NFOEC, March 2013
Follow-The-Sun Applications Source: Google Transparency Report 35
Virtual Machine Movement Floodlight Controller VM Cluster IBM V7000 Storage OpenDayLight Controller IBM G8264 OF Switch dual 10G dual 10G IBM G8264 OF Switch ADVA FSP 3000 Site A dual 10G ADVA FSP 3000 ADVA FSP 3000 dual 10G Storage Site C Site B Storage 36
Conclusion 37
Google s B4 Network From 30-40% to 98% Average Utilization! Typical 30-40% Utilization Source: S Jain, A Kumar, S Mandal, et al., B4: Experience with a Globally-Deployed Software Defined WAN, SIGCOMM 13, August 12 16, 2013, Hong Kong, China. 38
Thank you info@advaoptical.com IMPORTANT NOTICE The content of this presentation is strictly confidential. ADVA Optical Networking is the exclusive owner or licensee of the content, material, and information in this presentation. Any reproduction, publication or reprint, in whole or in part, is strictly prohibited. The information in this presentation may not be accurate, complete or up to date, and is provided without warranties or representations of any kind, either express or implied. ADVA Optical Networking shall not be responsible for and disclaims any liability for any loss or damages, including without limitation, direct, indirect, incidental, consequential and special damages, alleged to have been caused by or in connection with using and/or relying on the information contained in this presentation. Copyright for the entire content of this presentation: ADVA Optical Networking.