presentation11
TRANSCRIPT
1
OpenFlow History & Progress
Ethane
OpenFlow Consortium
OpenFlow 1.0 Spec
Open Networking Foundation
2007 2008 2009 2010 2011
Board: Deutsche Telekom, Facebook, Google (Chair), Microsoft, Verizon, Yahoo!Members: Big Switch Networks, Broadcom, Brocade, Ciena, Cisco, Citrix, Comcast, Dell, Ericsson, Extreme Networks, Force10 Networks, HP, IBM, IP Infusion, Juniper Networks, Marvell, Mellanox Technologies, Metaswitch Networks, NEC, Netgear, Netronome, Nicira Networks, Nokia Siemens Networks, NTT, Plexxi Inc., Riverbed Technology, Vello Systems, Vmware…Growing all the time
Nox, Open vSwitch
Academia
Silicon Vendors,Early adopters (NEC, Google)
Start-ups (Nicira, BigSwitch)
Telcos, System Vendors
OpenFlow 1.1In
tere
st
OpenFlow Usage Models1. Experiments at the flow level
User-defined routing protocols Network access control Network management Energy management VOIP mobility and handoff
2. Experiments at the packet level Slow: Controller handles packet processing Fast: Redirect flows through programmable hardware Modified routers, firewalls, NAT, congestion control…
3. Alternatives to IP Flow-table is Layer-2 based e.g. new naming and addressing schemes
Controller
PC
NetFPGA
Laboratory
Experiments at the packet level
OpenFlow-enabledCommercial Switch
FlowTableFlowTable
SecureChannelSecureChannel
NormalSoftware
NormalDatapath
Open Flow components
5
Available controllers and switches NOX (http://noxrepo.org/, GNU GPLv3)
Provides network-wide view of the topology C++ and Python modules make decisions
OpenVSwitch (http://openvswitch.org/, Apache 2) Soft-switch, replaces Linux bridge Designed to be used with VM's
Hardware switches: Quanta LB4G (Broadcom), NetFPGA
6
Analysis – The Potential• “SDN will open up networking”
– Do for networking what Linux did for the server – break the proprietary lock
– Vendors and DC Operators will be able to take control of their network without being limited to what switch vendors will give them
• Do-it-yourself rather than waiting 12 months to work it’s way through a vendor roadmap
– Create an open platform for innovation
• “Centralization of Control will yield better solutions”– Global view of data -> more efficient
• Processing will be done once (rather than in multiple devices per traditional distributed protocols)
– Smaller, simpler code
7
Analysis – The Potential
• “Workstations offer better platforms for processing large distributed datasets”– “Comp Science is years ahead of embedded in this respect” – e.g.
Hadoop– Better, richer, more productive programming environment– Larger, more accessible body of engineering skills
• “OpenFlow will result in lots of cheap switches!”– “White box” unbranded switches, possibly Open Source
• No vendor premium for the heavyweight software load• No vendor lock-in
– Small, cheap CPUs
FlowVisor Message Handling
OpenFlowFirmware
Data Path
AliceController
BobController
CathyController
FlowVisorOpenFlow
OpenFlow
Packet
Exception
Policy Check:Is this rule allowed?
Policy Check:Who controls this packet?
Full Line RateForwarding
Rule
Packet
10
Analysis – The Potential – Use Cases• ElasticTree: Reducing Energy in Data Center
Networks– Today data centers are provisioned for peak
traffic running at peak power– Improve the energy efficiency of a data center
network– Dynamically adjust network elements - links and
switches.– ElasticTree uses OpenFlow to measure traffic
statistics and control flow routes– Upto 60% savings demonstrated.
11
Analysis – The Potential – Some Use Cases
• Aster*x: Load-Balancing as a Network Primitive– Traditionally Load Balancing is done with an
expensive Box, sitting in the Data path.– Load Balancing is a just smart routing.– Transform an existing network into a distributed
load-balancing system.– Demonstrated one such OpenFlow-based load-
balancer called Aster*x– Load Balancing became a Control plane solution– http://www.youtube.com/watch?v=Sfqofxdk1gE
12
Analysis – The Potential – Some Use Cases
• Using All Wireless Networks Around Me– This demo shows how we can exploit all the
wireless networks around us to achieve better connectivity and hence better video streaming from a moving vehicle.
– simultaneous use of multiple wireless networks. – Uses OpenFlow Wireless-enabled WiFi and
WiMAX networks.
– http://www.youtube.com/watch?v=ov1DZYINg3Y
13
Analysis – The Challenges• “OpenFlow is too limited”
– How can you solve all networking problems with such a narrow set of primitives?
– All solutions will require lots of network services outside of OpenFlow in order to function, so does the “openness story” really hang together?
• “You cannot replace all the traditional switch/routing functions”– Need to maintain Controller connectivity across a network– Local processing required for HA/Fast failover– So will the switches really be any cheaper/simpler, or does OpenFlow support
become yet another switch feature?
• “SDN doesn’t scale”– Today switches do a lot of local processing (and need complex software and
big CPUs for a reason!) – they have a lot of dynamic, event-driven processing to-do
• Yes you can simplify this, but can you replace or export it?
– If you put all that up on a remote station, the both processing throughput and event latency will become big issues
14
Analysis – The Challenges• “Is it really that new? What can you do with
OpenFlow that we can’t already do with existing configuration methods?”
• “Solutions may move from being Switch vendor to Controller vendor dependent”– Where’s the interoperability?– Industry-hardened multi-vendor standards have
been available in traditional networks for years.
15
Predictions• SDN will supplement rather than completely replace traditional switch
features– Will still need much of traditional switching and routing for the foreseeable
future– See OpenFlow as a value-add feature
• SDN will create an innovation platform that will attract a lot of interesting solutions– OpenFlow Controllers will look more like OS’s – platforms not solutions– The Networking “App Store” will arrive!– However many solutions will require optional and proprietary features in the
switch
• SDN will create opportunities for silicon innovation– The richer the “instruction set”, the more powerful the solutions!
• Overall, this is a key trend that will happen, and will energize our industry
Thank You – Q&A
17
ConfigAPP1 to App2App3 to App1Web to DBEtc.
ConfigAPP1 to App2App3 to App1Web to DBEtc.
ConfigAPP1 to App2App3 to App1Web to DBEtc.
ConfigAPP1 to App2App3 to App1Web to DBEtc.
ConfigAPP1 to App2App3 to App1Web to DBEtc.
App1 App1
App3
App2 App2
App3
ConfigAPP1 to App2App3 to App1Web to DBEtc.
ConfigAPP1 to App2App3 to App1Web to DBEtc.
How OpenFlow works (Simplified)
Applications flows are preprovisioned throughout the network
Topology/application changes are reflected
APIs allow application to instruct network behavior
SDN/OpenFlow Customer Use Case 1- Trevela
18
Provider of a market-leading distributed fabric for global trading, risk analysis and e-commerce
Validated networking based upon OpenFlow, for predictable performance for complex and demanding applications
Fast Packet Forwarding using OpenFlow
Intelligent path selection
Line Rate Performance
Fast Convergence
Predictable performance
OpenFlow based traffic segregation
SDN/OpenFlow customer use case 2: Selerity
19
OF API
OF API OF API
OF API OF API OF API
Closed network elements forced the customer to a server to do the packet processing needed
Provider of ultralow-latency realtime financial information• Delivers machine-readable event data immediately as events are breaking which relies on
the fastest possible network performance
Performance impactNot an ideal solution
Used OpenFlow to get complete control of the network
Network based upon OpenFlow provided an order of magnitude (1000 times) better performance
SDN/OpenFlow Customer Use Case 3 – SPAN and Tap
20
DiagnosticComplianceMonitoring Auditing
Parallel network for diagnostics, compliance, auditing
Move flows from SPAN or TAP to OpenFlow switches
Cost-effective alternative to special-purpose devices
Open, standards-based cost-effective solution
Real World Deployments
FlowVisor
Alice Bob Cathy
IsolationPolicy
Scalable isolation domains and network slicing.
Example: Flowvisor
Network Virtualization
Inserting and managing network services, such as load balancing, firewall, IDS/IPS, QoS, etc.
Example: FlowScale
Platform for Network Services
Flexible mobility of virtual machines
Example: Stanford WAN VM Migration
Virtual Machine Management
CLOS Fabrics
Lower cost, high-performance networks
Example: non-blocking CLOS architectures
Simplified data vibility and traffic monitoring
Data Analysis / Monitoring Hybrid Clouds
Networks spanning public / private DCs
Example: Amazon VPC
21
OpenFlow as a strawman flow-based substrate
Our Approach1. Define the substrate
• OpenFlow is an open external API to a flow-table
• Version 1.0– Defined to be easy to add to existing hardware switches,
routers, APs, …– Timeframe: Now
• Version 2.0– OpenFlow-optimized hardware– General “flowspace”– Timeframe: 2011
OpenFlow Deployments• Stanford Deployments
– Wired: CS Gates building, EE CIS building, EE Packard building
– WiFi: 100 OpenFlow APs across SoE– WiMAX: OpenFlow service in SoE
• Other deployments– Internet2 (NetFPGA switches)– JGN2plus, Japan (NEC switches)– 10-15 research groups have switches
OpenFlow DeploymentsPlans in 2009-10
• Campus deployments– Lab + production use– “Enterprise GENI” (NSF/GPO)
• Backbone deployments– National research backbones– Research + Production use