Transcript G ANT Network Evolutionx

GEANT
Network Evolution
Mian Usman
Network Architect
NORDUnet Conference
September 2016
Helsinki
Agenda
The things causing us to think differently
Our challenges
What we have done
What we are doing
Our collaborations
What we want to achieve
On average human knowledge is doubling every 13 months
According to IBM the build out of the “internet of things” will
lead to the doubling of knowledge every 12 hours.
Human Brain Indexing Will Consume Several Billion Petabytes
In a lecture at Harvard University neuroscientist Jeff Lichtman,
who is attempting to map the human brain, has calculated that
several billion petabytes of data storage would be needed to
index the entire human brain.
We are seeing traffic volumes doubling every 18 months
The amount of data forecast to be generated and stored is
astounding. This will drive an exponential growth in network
traffic. IP backbone trunks will need to be 400Gbps by 2020 &
a terabit by 2025
3
It is not only traffic growth that is driving our thinking for how
our platforms (network + OSS) need to evolve
• Multi domain interoperation
•
•
•
•
Commonly supported technologies such as MPLS are key
Common OSS/BSS systems architecture
Common service offerings
Interoperate with all kinds of Service Providers & infrastructures
• Commercial
• R&E
• Common service offerings - In alignment with global partners; evaluated against commercial equivalents
• Technology & vendor agnostic services (hardware & software)
• Services built on standards based common components - Created out of a shared common OSS/BSS architecture
• Increasing complexity
• Services & functions move out of hardware into software
• Virtualised resources and functions
• Software Orchestration
• Sustainability
• More capacity
• Faster interfaces
• Operational complexity
4
This our challenge
• We
Exponential
traffic growthhere
adopt technology
+
• Need for more advanced services
+
We pay a premium
Forces us to adopt
technology early
• Greater operational complexity
5
New thinking is required to meet longer-term demands cost effectively
• Convergence of packet and optical networks
• Common packet /optical control plane
• Core network made simple to forward packets cost effectively
• Sophisticated services delivered on overlays - separation with core network
• Use of virtualised network functions
• Greater programmability
• Intelligent management plane(s)
• Vendor independence – separation of hardware and software
• Seamless interoperation of infrastructures
6
What have we done
Built a lab for evaluating new software & technologies
• Lab built to emulate production environment as closely as possible within realistic financial
constraints:
• Juniper MX routers and Infinera DTN-x, along with traffic and route generators, multicast
sources and receivers
• Configuration and code versions identical to production network
• BGP communities and network separation
• NetFlow v9
• Infinera OTSV
• PXM evaluation & Juniper DWDM
• Alien wave testing
• SDN evaluation and testing, Corsa DP6410; PICA-8 P-3922; Dell S4810-ON
• DTN testing
• Pre-staging and testing equipment before shipment
7
What have we done: Built a SDN Lab
Testing and development - PoC evaluation
• Equipment
• 2 x CORSA DP64x0
• 2 x PICA 8 Switches
• 2 x Dell S4810 Switches running
Cumulus Linux
• 1 x Dell Server for VMs
• Interconnected to the GÉANT Juniper MX
lab
8
What have we done
Successful implementation of new IAS AS to separate REN-to-REN from
REN-to-Internet traffic
NREN B
NREN A
Internet Access Service virtual routing and forwarding
(VRF) created – new AS established, REN-to-REN
traffic separated from REN-to-Internet traffic
• Cloud VRF added
GÉANT ren-I
ASN 21320
NREN C
Peers and upstream
moved to new
VRF/ASN
GÉANT ren-ren
ASN 20965
Peer/
upstream
NRENs have the ability to traffic engineer their route
propagation through the use of BGP communities,
with per-peer & per-PoP/IX granularity
Future creation of physical separation if required:
• REN to REN on 100Gbps
• REN to I on Nx10Gbps
9
What have we done
Sharing spectrum with NRENs
Identified routes where GÉANT and NRENs could benefit from infrastructure sharing
• Amsterdam – Hamburg in service
• Amsterdam – Paris in service Q4 2016
• Milan – Geneva
• Madrid – Lisbon
• Paris – Lisbon
• Marseille – Milan
• Marseille – Madrid
• Budapest Zagreb
10
What have we done
Packet optical integration
IP-over-OTN architecture is now deployed using
Infinera PXM cards
JDWDM implementation is also being
considered under lab testing and trialed in
future
We
can did
now
disaggregate
100Gbps
IP trunks
follow
the fibre to reduce
the interface
without
of the nx10G
performance
issues and
number any
of 100Gbps
interfaces
& cost
create multiple direct routes that do not follow the
fibre
11
What are we doing: Transport and multi-layer SDN
Software control of packet and optical layers
SDN
controller
OTSv
L2 SDN
switch
GÉANT
Router
L2 SDN
switch
L2 SDN
switch
100GbE
100GbE
100GbE
GÉANT WDM/OTN
GÉANT
Router
ODUflex and OTSv
used to dynamically
instatiate these links
100GbE
NREN
N x 10GbE
N x 10GbE
Legacy ODU4/2e links to
routers via TIM ports
IP trunks. ODUflex used to
manage bandwidth between
PXM cards.
These are static (not OTS controlled)
100GbE
NREN
GÉANT
Router
PXM cards support
termination of EVPL
circuits on vlans
GÉANT
Router
12
What are we doing: Developing SDN solutions for different domains
New features, future services
• Solutions for the carrier backbone
• Software-Defined Internet Exchange
(SDX) at Layer 3 (SDN IP) and Layer 2
• Transport SDN
• SDN-based Bandwidth on Demand
(BoD) with advanced path computation
capabilities
• Solution for intra-campus
networking
• Using an SDN
controller to stitch
flowspaces to VLANs
within the campus
• Solution for intra-DC
networking
• Overlay networking
and L2 gateway
solutions
13
What are we doing: Developing a much needed service - SDN-enabled BoD
SDN-powered, guaranteed-capacity circuits on demand
TS
GUI
NSA
NSI-CS
IDM
NSI-CS
NSA
Rest
API
Rest
API
DynPaC
DynPaC
DynPaC
ONOS
ONOS
Technology
Proxy
OpenFlow
SW01
Pica8
OpenFlow
SW02
Pica8
Junos
Space
SW08
OVS
SW09
OVS
Domain 3
Domain 2
VM
SW05
CORSA
Domain 1
SW06
CORSA
Cambridge Lab SDN
Juniper
MX
Juniper
MX
Juniper
MX
Juniper
MX
GÉANT-BoD non-SDN
SW07
OVS
SW11
OVS
SW10
OVS
• Vendor-agnostic implementation
• Efficient use of the network capacity
• Flow relocation
• Flow disaggregation
• Resiliency in case of a link failure
with quick recovery times
• Pre-computed backup paths
VM
AMRES SDN
14
Vendor collaboration
GÉANT-tailored vendor developments
• Delivering L2-L3 SDN data plane
functionality by leveraging CORSA
hardware features
• Support for 10/40/100GE,
transceiver agnostic, enhanced QoS
(Deep buffers), OF 1.3 compliant,
multi-table pipeline
• Co-developing a solution for managing
OTN (PXM) cards using ONOS
• Leveraging OTSv
• Using REST to request an Ethernet virtual
private line (EVPL) service between 100Gbps
ports
• GÉANT-tailored multi-table pipeline
• Metering
• ONOS driver
15
Collaboration with ON.Lab
GÉANT contributing to global community open-source controller
• Presentation of GÉANT service definitions
• Specialised presentations on SDN controller requirements
• Comparison of ONOS roadmap with GEANT roadmap
• 10s of GEANT commits to the ONOS core codebase:
• https://github.com/opennetworkinglab/onos/commits
• https://github.com/opennetworkinglab/onos-app-samples/commits
• Most of them reviewed and merged
• Flow-objective framework capability not delivered as promised:
• GEANT developer hosted by ON.Lab to implement and integrate the required
functionality
• Three-way collaboration with Corsa on the SDX L2/L3 cases
• Three-way collaboration with Infinera on the Transport SDN case
16
What we want to achieve
Put S DRIVEN N into operation – PRODUCTION SDN
• Determine the right way to separate control, data and management planes
• Power from vendors into the hands of Users and Operators
• Integration between our packet and optical platforms
• Enable single control across these layers – use expensive infrastructure more efficiently
• Router replacement
• Alternative to existing vendor solutions
• Seamless interoperation of infrastructures
• Truly multi domain
• Move services into software
• Become vendor agnostic
• Cost effectively run the GEANT network in the future
17
Thank you and
any questions?