1) OpenFlow: One of a number of possible SDN approaches
2) SDN: Empowers the operator with reduced OPEX and CAPEX
3) PCE: Solves some real-world carrier problems
2. Software Defined Networking (SDN) 2
Smart
OSS/BSS Control
Cloud
Network Equipment Dumb Network
Traditional SDN
Key elements of SDN are
Intelligence moves from NEs to control cloud
Physical separation of control plane / data plane
3. What Does SDN Offer? 3
Empowering the operator
Reduced OPEX
Simplify operation by more powerful abstraction of the network
Optimize resource usage through centralized data and algorithms
Simplify process for upgrading network software
Reduced CAPEX
Cloud offers better price / performance for CPU power
New possibilities
Specific per-operator traffic optimizations
More dynamic network topologies
Network virtualization
Facilitate innovation
Up till now this has been the domain of the NE vendor
4. OpenFlow – One Form of SDN 4
Smart
Control OpenFlow formalizes this interface
Cloud
For each received packet
• Match fields in packet header
• Perform specified action
Dumb Network
Standard control/data plane interface
Stanford / Open Networking Foundation
OpenFlow v1.2 recently finalized
Inter-operable
Offers considerable flexibility
OpenFlow paradigm is complex
So cloud software tools to abstract it are critical
5. OpenFlow Applications 5
Began life in the research community
Prototyping new networking schemes
Evolving for more general deployment
OpenFlow paradigm has wide applicability
Operators and carriers
Data centres
Organizations with unique networking requirements
Ethernet, IP (and maybe MPLS)
It requires new software tools
OpenFlow constructs are too low level for users
Success fundamentally depends on these tools
6. Path Computation Elements – also SDN 6
Path
Computation PCE also formalizes this interface
Elements
NEs operate much as today
• Signaling function resides on NEs
• Routing function resides in cloud
Semi-smart
Network
Define a “route-query” interface
IETF standard for PCE protocol (“PCEP”)
RFCs 4655 5088 5089 5394 5440 5441 5520 5521 5541 5557 6007
Architecture RFCs 6 years old, Specification RFCs 3 years old
Simple network paradigm
“How should this (G)MPLS traffic flow be routed?”
NEs query the cloud (opposite way round from OpenFlow)
7. PCE for Inter-Area TE Routing 7
Client/server
PCE
Route Query
Server See this live
at the EANTC
TE Routing
Updates Interop Event!
Ingress LER Egress LER
LSP
Area A Area B
PCE server can be in the cloud
Provides CPU power for potentially massive TE database
Operator can interact with it
And therefore influence routing policy
NEs retain signaling intelligence
The nuts and bolts that make the network function
8. More Advanced PCE Capabilities 8
PCE Server PCE Server
(domain A) (domain B)
Inter-domain route query
Ingress LER Egress LER
LSP
Domain A Domain B
Multiple PCE servers can used
Keeps carrier’s internal topology private
Domain A sees only an encrypted key to Domain B path
Key carried in signaling and expanded once in Domain B
Auto-discovery of PCE servers
Redundant PCE servers
9. PCE for WDM/OTN 9
Routing is a major user of CPU power
Same colour needed end-to-end
Full RWA algorithm can require huge processing power
NEs may have restricted connectivity
Connectivity matrix adds routing complexity
Optical impairments
PCE offers a way forward
Lower cost CPU capacity for complex algorithms
Simpler software upgrades as the algorithms evolve
Carriers visibility into routing decisions
And crucially control the routing policy or algorithm
PCE can be integrated / co-located with OSS
10. PCE for Multi-Layer Networking 10
Client/server
PCE
Route Query
Server
TE Routing
Updates Cloud software
auto-provisioning tool
Ingress LER Egress LER
Out of capacity
Add another lambda
And use it when ready
PCE server triggers auto-provisioning tool
Delay LSP setup
Provision an extra wavelength
When ready, tells ingress LER to use it
Tool can include operator provisioning rules
11. Key Take-Aways 11
OpenFlow
One of a number of possible SDN approaches
Getting lots of attention, but still early days
Revolutionary and disruptive
Likely first applications
Ethernet and IP
Networking “Islands” e.g. data centres
PCE
Solves some real-world carrier problems
Well established IETF standards
See PCE at the
Evolutionary EANTC stand
Likely first applications
Transport networks e.g. WDM, OTN, MPLS-TP
12. 12
THANK YOU
Pat.Moore@Metaswitch.com
http://network-technologies.metaswitch.com/