SDN: New Approach to Networking
Guru Parulkar
parulkar@stanford.edu
http://onrc.net/
1
Team at Stanford
Nick McKeown, Guido Appenzeller, Nick Bastin, David Erickson, Glen Gibb, Nikhil
Handigol, Brandon Heller, TY Huang, Peyman Kazemian, Masayoshi Kobayashi, Jad
Naous, Johan van Reijendam, Srini Seetharaman, Rob Sherwood, Dan Talayco, Paul
Weissman, Tatsuya Yabe, KK Yap, Yiannis Yiakoumis and many more.
With Scott Shenker and team at Berkeley and Martin Casado at Nicira
Problem with Internet Infrastructure?
Routing, management, mobility management,
access control, VPNs, …
Feature
Feature
Operating
System
Specialized Packet
Forwarding Hardware
Million of lines
of source code
6000+ RFCs
Billions of gates
Bloated
Power Hungry
Vertically integrated, complex, closed, proprietary
Not suitable for experimental ideas
Not good for network owners & users; Not good for researchers.
Problem: No Abstractions for Control Plane
• Addition of a new function to the network
–
Highly complex distributed system problem
• Networks too difficult to program and to reason about
– no good abstractions and interfaces
Distributed
Network
Functions
OS
OS
Forwarding
Forwarding
Router/Switch/Appliance
Router/Switch/Appliance
State Distribution
Mechanism
OS
Forwarding
Router/Switch/Appliance
Not good for even network vendors
4
Cloud Provider Today
• Assign VMs to any server without considering L2/L3 scalability
• Support multiple tenants with lots of VMs interconnected
• Support each tenant with its own customized network for VMs:
topology, bandwidth, security, load-balancing
And everything is dynamic
Tenant-A
Load balancers
Tenant-B
Tenant-C
Firewalls
IDS’s
?
Network is the long pole
VM VM VM
VM VM VM
VM
VM VM VM
VM VM VM
VM VM VM
VM
VM VM VM
VM VM VM
VM VM
VM VM
VM VM VM
VM VM VM
VM
VM
VM VM VM
VM VM VM5
Other Descriptions of the Problem
• Network requires device oriented low level config and
management
• Network is not programmable plug-n-play component
– for private and public cloud infrastructure and services
Software-Defined Network
with Key Abstractions in the Control Plane
Well-defined open API
Routing
TE
Mobility
Network Map
Abstraction
Network OS
Open interface (OpenFlow)
to Forwarding Abstraction:
L1/L2/L3
Packet
Forwarding
Programmable
Basestation
Packet
Forwarding
Packet
Forwarding
Separation of
Data and Control
Plane
Software Defined Network
with Virtualization
Abstract Network Model
Network
ControlVirtualization
Program
Global Network View
Network OS
Innovation/Research Enabled?
9
Nation-wide SDN Infrastructure
Part of NSF’s GENI
SDN Reference
Example Research Enabled
•
•
•
•
•
•
•
•
•
•
Data center: energy conservation, routing, and management
Seamless use of diverse wireless networks
Network based load balancing
Packet/circuit convergence, traffic engineering
Simpler control plane for converged packet/circuit MPLS nets
Slicing and remote control/management of home networks
Distributed snap shot of VMs (by DOCOMO researchers)
Inter-domain routing with pathlets (by UIUC)
Redundant traffic elimination [for CDNs] (by Univ of Wisconsin)
And many more …
200+ OpenFlow/SDN deployments around the world!!
Example Demonstrations
Onrc.stanford.edu/videos.html
13
Stanford/Berkeley SDN Activities
With Partners
Platform
Development
Deployment
Demo
SIGCOMM08
Ethane
VM Migration
(Best Demo)
Over 68 countries
(Europe, Japan, China, Korea,
Brazil, etc.)
US R&E Community
GENI: 8 Universities + Internet2 + NLR
Many other campuses
Stanford University
~45 switch/APs ~25user
In McKeown Group
GENI software suite
Tools
Reference Switch
2008
CIS/EE Building
Production Network
Expedient/Opt-in Manager/FOAM
Virtualization
Network OS
Interop
2011
Trans-Pacific SDN Concept Baby GENI Nation Wide GENI “The OpenFlow Show”
VM Migration (Best Demo)
– IT World
Other countries
OpenFlow Spec
2007
GEC3
GEC9
SIGCOMM09 GEC6
FlowVisor (Java)
FlowVisor
NOX
Test Suite
NetFPGA
v0.8.9
2009
SNAC
Beacon
oftrace
Software
Measurement tools
+Broadcom
v1.0
2010
v1.1
2011
Mininet
Not Just Research
All parts of networking industry
embracing OpenFlow and SDN
65+ companies joined ONF in a year
65+ companies from across the industry
“HP Simplifies Networking with Broadest Choice of
OpenFlow-enabled Switches, Feb. 2, 2012”
“IBM & NEC Team Up to Enable Industry Innovators Tervela
& Selerity to Transform their Networks w/ OpenFlow”
“Nicira Comes Out Of Stealth Mode With “GameChanging” Network Virtualization Platform”
“NEC Adds 10/40GbE OpenFlow Switch to AwardWinning ProgrammableFlow Product Line.”
“Pica8 launches 3920.”
“Brocade Leads OpenFlow Adoption to Accelerate Network
Virtualization and Cloud Application Development.”
SDN Exhibits by 20+ Companies at ONS-12
Why does the industry care?
21
Why Does Verizon Care?
Stu Elby @Open Networking Summit
http://OpenNetSummit.org/
Software deļ¬ned network
enables inexpensive feature
insertion for new services and
revenue uplift.
ARPU(PMO)
Pmargin
Fmargin
$
Unacceptably high cost
escalation results in a
non-sustainable business
case.
The cost of PMO needs
to be matched to the
characteristics of revenue
growth to support
consistent margin.
Target FMO costs
Using COTS hardware
following mass market
cost curves lowers
equipment expenditures
Cost of PMO
The cost of transition
must not be prohibitive
t
© 2011 Verizon. All Rights Reserved.
Confidential and proprietary materials for authorized Verizon personnel and outside agencies only. Use, disclosure or distribution of this material is not permitted to any unauthorized persons or third parties except by written agreement.
22
Carrier Networks Today
They want to offer
• VPN Services
• Traffic Engineering
And other value added services
Go to vendors and ask them for solutions:
State Distribution
Mechanisms
Distributed Network Functions
each with their own
State Distribution Mechanisms
Vendors find it hard to develop solutions
Providers find it too slow and too expensive
to deploy and operate
RouterProviders depend too much on vendors
Operating
Distributed
Network
Functions
PE
Label
Distrib
ution
E-BGP
learned
Route
Advert
VPNIPv4
Route
Advert
TE
Label
Distrib
ution
IGPRoute
Advert,
LinkState
LDP
I-BGP
+ RR
MPBGP
RSVPTE
OSPF
v2
System
23
Carrier Networks (MPLS) with SDN
~4K vs. ~600k lines of code
Services / Network
Applications
(3)
Simpler
Control Plane
(2)
TE 2.0
Routing
Discovery
VPN
Label
Distribution
Recovery
NETWORK OPERATING SYSTEM
Vendor-agnostic Open Interface
Simpler
Cheaper
Multi-Vendor
Data Plane
(1)
PUSH
SWAP
POP
24
Video of a
Demonstration
showing MPLS-TE service with
the Map Abstraction
Onrc.stanford.edu/videos.html
25
IP Network
Transport Network
26
Converged Packet/Circuit with SDN
2. Common Map
Abstraction
Unified
Control
Plane
Network OS
Interface: OpenFlow Protocol
1. Common Flow
Abstraction
Packet &
Circuit
Switches
Converged Network
Converged Packet/Circuit with SDN
Routing
TE
QoS
2. Common Map
Abstraction
Unified
Control
Plane
Network OS
Interface: OpenFlow Protocol
1. Common Flow
Abstraction
Packet &
Circuit
Switches
Converged Network
Application across
packet and circuits
Prototype
Packet switches
NOX
Hybrid Packet-Circuit Switches
29
Video of a
Demonstration
of network application
on Prototype
Onrc.stanford.edu/videos.html
30
Capex Savings with Converged SDN
Architecture:
1. Replace BRs with
hybrid MPLS-OTN
(packet-optical)
switches
2. Full mesh topology
of variable
bandwidth circuits
used for recovery
and BoD
3. Use of SDN based
unified control
plane
67%
Urs Hölzle, SVP, Google at ONS 2012
http://OpenNetSummit.org/
•
•
•
•
•
•
•
applica=on%development%
%
NTT To Offer Global IaaS Leveraging OpenFlow
Technology
By Chad Berndtson, CRN
June 11, 2012 9:46 AM ET
Enterprise Networking
Enterprise network operators want..
• Firewall and access control
• Delegate management to departments
• Lots of VLANs that stretch across buildings
• By-pass bottlenecks/check points for specific applications
• Host web services with load balancing
• Easy guest wireless access with security
And more
How do they do it today?
35
Enterprise Network: Today’s solution
Firewall
Load balancer
ACL
IDS
ACL
ACL
Firewall
IDS
Load balancer
ACL
ACL
• Proliferation of appliances
• Increased management complexity
- Device oriented management
- Each device type has its own management
• High Capex, high Opex
• Too much reliance on vendors
36
Enterprise Network with SDN
And you can even delegate control to someone else
Financial Department
Research Labs
Centralized
Control Plane
Load
Balancing
Access
Control
IDS
Policy
Routing
NETWORK
OPERATING SYSTEM
Policy
Routing
IDS
NETWORK
OS
Vender-agnostic
Open Interface
Firewall
Load balancer
Simple, Cheaper
Multi-vendor
Data Plane
ACL
IDS
ACL
ACL
Firewall
IDS
Access
Control
NETWORK OS
Load balancer
ACL
ACL
What Stanford IT and others are exploring …
Data Center Provider Today
• Assign VMs to any server without considering L2/L3 scalability
• Support multiple tenants with lots of VMs interconnected
• Support each tenant with its own customized network for VMs:
topology, bandwidth, security, load-balancing
And everything is dynamic
Tenant-A
Tenant-B
Load balancers
?
Firewalls
IDS’s
Tenant-C
VM VM VM
VM VM VM
VM
VM VM VM
VM VM VM
VM VM VM
VM
VM VM VM
VM VM VM
VM VM
VM VM
VM VM VM
VM VM VM
VM
VM
VM VM VM
38
VM VM VM
Cloud Provider with SDN
Tenant-B
Tenant-A
Load
Balancing
Firewall
NETWORK OS
VM
VM
VM
VM
VM
VM
VM
VM
Load
Balancing
IDS
Tenant-C
Load
Balancing
NETWORK OS
Firewall
IDS
NETWORK OS
VM
VM
VM
VM
VM VM VM VM VM VM
VM VM VMVM
VM VM VM
VM VM VM
VM
VM VM VM
VM VM VM
VM VM VM
VM
VM
VM VM VM
VM VM VM
VM VM
VM VM
VM VM VM
VM VM VM
VM
VM
VM VM VM
VM VM VM
VM VM VMVM
39
Berkeley and Stanford establish Open
Networking Research Center (onrc.net)
with 12 founding members
Scope of Activities
Abstractions
Big Connector
Virtual Network
Feature
Feature
Feature
Big Connector
Programming System/Tools
Virtual Network
Network Map
Network HyperVisor
Logical Map of Network
Network OS
Flow Space
OF
Switch
Scalability
Reliability
Security
OF
Switch
Mininet
& Tools
Systems
Open
Radio
OF
Switch
OF
Switch
Xlities
Network Network
FlowVisor
OS-wireless
OF
OS
OpenRadio
Switch
Prog
Systems
Scope of Activities Cont.
Domains of Use
Enterprise Networks
Abstractions
Big Connector
Virtual Network
Datacenter Networks
OF
Switch
Network Map
Flow Space
Systems
Scalability
Reliability
OF
Switch
Mininet
& Tools
Open
Radio
Network Network FlowVisor
OS-wireless
OS
Service Provider Networks
Prog
Systems
Security
Cellular Networks
Xlities
Home Networks
ONRC: Open Networking Research Center
Berkeley Stanford
Scott Shenker
Nick McKeown
Guru Parulkar
Sachin Katti
Open Network Lab
Exec Director: Guru
VP Eng: Bill Snow
12-15 Engineers/Tech Leads
PhD/Postdocs
Research
Open Source SDN Stack for
Growing Community
Develop, Deploy, Support
Open Source SDN
ON.LAB is Recruiting
Motivated talented software
engineers who are passionate
about
• working with SDN leaders
• developing open source SDN
stack
• changing the world of
networking and also
• working at a very nice fun
facility in Palo Alto down
town
Interested? Visit http://onlab.us/
“OpenFlow and Software Defined Networking
(SDN) are not only here to stay, but they will
define the future of networking.”
Network World, 10/18/2011