SDN in GÉANT provides three key points: 1) GÉANT has implemented two generations of SDN testbeds to enable researchers to test novel network ideas. 2) The Network Service Interface (NSI) protocol solves the multi-domain SDN service challenge by enabling flexible, multi-domain service chaining across network domains. 3) Work is ongoing to integrate NSI into the SDN environment by adding OpenFlow constructs to the NSI topology descriptions to bridge intra-domain OpenFlow operations with inter-domain NSI provisioning.

1. SDN in GÉANT
Guy Roberts, DANTE
SDN conference, 17 June 2014

2. 2Connect | Communicate | Collaborate
Agenda
• SDN in European R&E: what’s SDN good for?
• SDN research in the GÉANT project
• The NSI protocol: solving the multi-domain problem
• Building a global circuit service based on NSI
• NSI as an SDN building block

3. 3Connect | Communicate | Collaborate
“GÉANT” is many faceted…
• A European network backbone operated by DANTE
• A network service area backbone + NRENs
• A consortium (of NRENs, DANTE & TERENA)
• An EC co-funded FP7 project (GN3Plus)

4. 4Connect | Communicate | Collaborate
GÉANT: The European internet backbone
• GÉANT is co-funded by Europe’s NRENs and the European Commission (EC)
under the Seventh Framework Programme (FP7)
• Project Partners are 38 European NRENs, NORDUnet, TERENA and DANTE
as Co-ordinator
• 150 FTEs’ annual effort (> 350 individuals) working in GÉANT
across Europe
25 European
POPs
12,000km of dark
fibre on 18 routes
50,000km network
infrastructure on
44 routes
Diversified footprint
• Serves 40 million users
• 8,000 institutions
• Across 40 European
countries

5. 5Connect | Communicate | Collaborate
SDN drivers in R&E networks
What is of interest in Research Networking?
• End-to-end performance
• Supporting large science flows
• Giving control over the network to researchers
• Solving the multi-domain service challenge
What is NOT interesting?
• Cost reduction
• Billing

6. 6Connect | Communicate | Collaborate
GN3Plus Project - activities
SA1: Core Backbone Services
SA3: Network Service Delivery
SA5: Application Services
SA2: Testbeds as a Service
SA4: Network Support Services
SA7: Support to Clouds
JRA1: Network
Architectures for
Horizon 2020
JRA2: Technology
Testing for Specific
Service
Applications
JRA3: Identity &
Trust Technologies
for GÉANT Services
SA6: Service Management & Operation
NA1:
Management
NA2:
Communications
& Promotion
NA3: Status &
Trends
NA4:
International &
Business Devpt

7. 7Connect | Communicate | Collaborate
SA2: Testbeds
Two generations of SDN testbed
• Generation 1
GÉANT Open Flow Facility
SDN focused testbed based on Xen hypervisor, a full
mesh of Open vSwitches coordinated using the
Ofelia Control Framework
• Generation 2
Testbed as a Service
Low level Testbed resources are allocated via UI.
Open Stack is used for the VMs and OF enabled
switches are interconnected via NSI enabled BoD
service.

8. 8Connect | Communicate | Collaborate
Gen 1: GÉANT OpenFlow Facility
User VMs
XEN
GÉANT
Frankfurt PoP
OpenFlow Protocol
Dedicated User
Controller A
Internet
User VMs
XEN
GOCF
Data Plane
User VMs
XEN
Controller
(other)
User VMs
XEN
User VMs
XEN
OpenFlow switch
GOCF
FlowVisorGeneric-
Shared
controller
Dedicated User
Controller B
Vienna
Zagreb
London
Amsterdam

9. 9Connect | Communicate | Collaborate
Gen 2: Testbed as a Service
RM
Resource A
port p0,
p1;
Resource B
port
out1, out2;
Adjacency
B/out1==A/p0;
Researcher has a
brilliant idea
A C
B
Ethernet Switch
“B”
VLAN “L1”
Testbed “Alpha” Description
X86 Server
“C”Virtual
Circuit
“L3”
VLAN
“L2”Virtual Machine
“A”
User logs in, and builds
a testbed description
via a web GUI frontend
to their Testbed Control
Agent
Resource Manager
Allocates resources and sets up the testbed
control plane
Network testbed concept
to test novel idea
TCA
Testbed Description
Doc fed to RM
Testbed is activated and user
controls it via the TCA
TCA

10. 10Connect | Communicate | Collaborate
Gen 2: Testbed as a Service
RM
Resource A
port p0,
p1;
Resource B
port
out1, out2;
Adjacency
B/out1==A/p0;
Researcher has a
brilliant idea
A C
B
Ethernet Switch
“B”
VLAN “L1”
Testbed “Alpha” Description
X86 Server
“C”Virtual
Circuit
“L3”
VLAN
“L2”Virtual Machine
“A”
User logs in, and builds
a testbed description
via a web GUI frontend
to their Testbed Control
Agent
Resource Manager
Allocates resources and sets up the testbed
control plane
Network testbed concept
to test novel idea
TCA
Testbed Description
Doc fed to RM
Testbed is activated and user
controls it via the TCA
TCA

11. 11Connect | Communicate | Collaborate
ARES
Employing cloud techniques to move Genome data
• The ARES project is leveraging cloud services for genome content distribution
• Individual genome sequencing for everyone is imminent, this will result in huge
amounts of data - current network architectures won’t scale.

12. 12Connect | Communicate | Collaborate
ARES: Advanced Networking for EU
Genomic Research
• A range of CDN services need to be supported, such as:
• Minimum delay CDN services for handling urgent situations.
• Short delay CDN services for handling less urgent situations.
• Balanced network load CDN services for handling all other situations.
• Management of virtualization services using NetServ (NSF funded
programmable router architecture).
• NSIS signalled caching allows data to be accessed via a cloud-like interface
• Virtualization through OpenStack based IaaS allows aggregation of computing
resources and storage.
• Use of NSIS NSLP protocol for discovering bottlenecks & disjoint paths
Research by University of Perugia and Polo GGB and funded by GÉANT Open Calls

13. 13Connect | Communicate | Collaborate
Autonomic OpenFlow (AUTOFLOW)
• AUTOFLOW is applies autonomic network management (ANM) to the SDN
environment. Uses ANM Control Loops (ACLs).
• Focus of work is on self-configuration and self-healing properties of ANM, in
order to recover from network failures and optimize network performance
• Goal is to steer the network’s operation according to operator’s policies (energy
efficiency, load balancing, service performance etc)
• Employment of reinforcement learning and an agent system per physical
network that is responsible for customizing its resources
• SDN enables the centralization of the routing - inter-AS optimization and allows
load balancing algorithms to be applied
• Congestion prediction, based on the past experience of the network.
• learning scheme uses Self-Organizing Map (SOM) which is a artificial neural
network
Research by University of Perugia and Polo GGB and funded by GÉANT Open Calls

14. 14Connect | Communicate | Collaborate
NSI Fundamental Design Principles (1/3)
NRM
1. “Network Service Interface” is a framework for inter-domain
service coordination
NSA
NSA
Network Services
Agent (NSA)
Requester
Agent (RA)
Provider
Agent (PA)
Network
Services
Interface
Network Resource Manager
(NRM)
NSI Network Service Domain
14
Supports
advance
reservations
Examples:
• Connection Service (NSI-CS)
• Document Distribution Service (NSI-DDS)
• Monitoring Service
• Protection Service
• Verification Service
• Etc.

15. 15Connect | Communicate | Collaborate
2. Designed for flexible, multi-domain, service chaining
Domain CDomain BDomain A
Tree NSI Topology
Supports Tree and Chain model
of service chaining
Fits in well with Cloud/Compute model of
provisioning as well as Network/GMPLS model
Domain CDomain BDomain A
NSA
Aggregator NSA
Chain NSI Topology
NSA
NSA NSANSA NSA NSA NSANSA
ultimate RA
ultimate PA uPA uPA
uRA
Aggregator/
uPA
Aggregator/
uPA
Aggregator/
uPA
NSI Fundamental Design Principles (2/3)

16. 16Connect | Communicate | Collaborate
NSI Fundamental Design Principles (3/3)
3. Principles of Abstraction applied – to network layers,
technologies and domains
Service Termination
Points (STPs)
Service Demarcation
Points (SDPs)
Are both abstract
and technology
independent

17. 17Connect | Communicate | Collaborate
NSI compliant dynamic circuit
implementations
• AutoBAHN – GÉANT (Poznan, PL)
• BoD - SURFnet (Amsterdam, NL)
• DynamicKL – KISTI (Daejeon, KR)
• G-LAMBDA-A - AIST (Tsukuba, JP)
• G-LAMBDA-K – KDDI Labs (Fujimino, JP)
• OpenNSA – NORDUnet (Copenhagen, DK)
• OSCARS – ESnet (Berkeley, US)
• NSI Documents: https://redmine.ogf.org/dmsf/nsi-wg?folder_id=6526
• NSI Co-Chairs
Guy Roberts <guy.roberts@dante.net>
Chin Guok <chin@es.net>
Tomohiro Kudoh <t.kudoh@aist.go.jp>

18. 18Connect | Communicate | Collaborate
How does NSI fit into SDN?
NSI for multi-domain path negotiation
uRA
uPA
A B
uPA
C D
uPA
E F
Host Host
OF
FlowVisor
OF
FlowVisorNRM
NSI
NSI NSI
NSI NSI NSI
AG AG AG

19. 19Connect | Communicate | Collaborate
Infinera’s Transport SDN Models
SDN Controller
Explicit (Direct) Model
OpenFlow
POTN
OTS
P-OTN
LSR
LSR ENET
OTS OTS
MPLS LSR
Ethernet
LSR
OTS
LSR
OTS
POTN
OTS
POTN
OTS
POTN
OTS OTS ENET
OTS
ENET
OTS
ENET
OTS
POTN
POTN
POTN
OTS
P-OTNLSR
LSR
LSR
LSR ENET
ENET
ENET
ENET
OTS OTS
OTS
MPLS LSR
Ethernet
Implicit (Indirect) Model
OpenFlow
SDN Controller
• Network abstraction per domain
• Multi-domain orchestration
• Leverage existing control plane
POTN
OTS
• Centralization of all network control
• Individually controlled NE’s
• Hop-by-hop provisioning
MPLS CP GMPLS CP

20. 20Connect | Communicate | Collaborate
SDN and NSI integration
• GÉANT sees the NSI protocol as a key component in delivering
multi-domain SDN services in the R&E networks.
• Work is ongoing in the MOTE GÉANT Open Call to integrate NSI
into the SDN environment.
• Aims to add Open Flow constructs to the NML topology
description.
• This will bridge the intra-domain operations of OpenFlow with the
inter-domain provisioning in the NSI.

21. 21Connect | Communicate | Collaborate
www.geant.net
www.twitter.com/GEANTnews | www.facebook.com/GEANTnetwork | www.youtube.com/GEANTtv
Connect | Communicate | Collaborate
Thank you!