1. Flow Abstraction End to End Flow L4: TCP src/dst port L3: IP
src/dst addr, IP proto L2.5: L2: Flow Identifiers Common Dest Flow
L4: L3: IP dst prefix for China L2.5: L2: 4
Slide 5
1. Flow Abstraction Classification of packets that have a
logical association Action & Maintaining Flow State Flow based
Accounting & Resource Management What is a Flow? L4: L3: IP src
prefix for branch L2.5: L2: Flow Identifiers Common Src Flow L4:
TCP dst port 80 L3: IP proto L2.5: L2: MAC src Web traffic from a
Handset L4: L3: L2.5: MPLS Label ID L2: All packets between 2
routers 5
Forwarding Hardware App Forwarding Hardware App Forwarding
Hardware App Forwarding Hardware App Forwarding Hardware Operating
System Operating System Operating System Operating System Operating
System Operating System Operating System Operating System Operating
System Operating System App Network Operating System App What is
SDN? 13
Slide 14
App Forwarding Hardware App Forwarding Hardware Network
Operating System A.Open interface to hardware C. Well-defined open
API B. Network-wide Operating System (extensible, built to scale)
The Software-defined Network The Software Defined Network 14 1.Flow
Abstraction 2. Map Abstraction
Slide 15
Outline 2 Abstractions & SDN 2 Research Examples 1.MPLS
2.Packet-Optical
Slide 16
MPLS and Flows Label Switch Router (LSR) LSPs Incoming packets
Classification Into FECs Flow state in Head-end LER MPLS network IP
networkLabel Switched Path (LSP) Label Edge Router (LER) MPLS has
Flow Abstraction
Slide 17
Why do Service Providers use MPLS? Really about 2 services MPLS
Services MPLS VPNs MPLS - TE Motivation Highly profitable No easy
way Older ways not used Motivation Deterministic Behavior Efficient
Resource Utilization Older ways not used
Introducing Map Abstraction in MPLS OSPF-TE RSVP-TE LDP I-BGP
OpenFlow NETWORK OPERATING SYSTEM Routing Routing Discovery
Discovery Label Distribution Label Distribution Recovery Recovery
TE TE LMP MP-BGP PUSH Simpler Data Plane Simpler Control Plane
Services Network Applications Label Switched Path (LSP)
Slide 20
Open vSwitch with standard MPLS data plane Open vSwitch with
standard MPLS data plane Prototype System Network Operating System
(NOX) GUI (Envi) showing real-time network state GUI (Envi) showing
real-time network state Open vSwitch (with MPLS) Open vSwitch (with
MPLS) Open vSwitch (with MPLS) Open vSwitch (with MPLS) Open
vSwitch (with MPLS) Open vSwitch (with MPLS) Open vSwitch (with
MPLS) Open vSwitch (with MPLS) Open vSwitch (with MPLS) Open
vSwitch (with MPLS) Open vSwitch (with MPLS) Open vSwitch (with
MPLS) Open vSwitch (with MPLS) Open vSwitch (with MPLS) Open
vSwitch (with MPLS) Open vSwitch (with MPLS) Open vSwitch (with
MPLS) Open vSwitch (with MPLS) Open vSwitch with standard MPLS data
plane Open vSwitch with standard MPLS data plane OpenFlow MPLS GUI
MPLS API MPLS Stats CSPF Routing MPLS-TE Auto route; Auto bandwidth
Traffic aware LSPs; Priorities TE-LSP configuration Mininet
Environment
Slide 21
Video of a Demonstration showing MPLS-TE service with the Map
Abstraction 21
http://www.youtube.com/watch?v=EpttFVKUrzs&feature=channel_video_title
Slide 22
TE-LSP Features 1.Auto-route 2.Auto-bandwidth 3. Priorities
4.Load-share 5.Diffserv aware Traffic Engineering (DS-TE) 6.MPLS
FRR 7.Explicit Routes 8.Re-optimization timers 4000 lines of code
Vs. 80,000 + ?
Slide 23
Outline 2 Abstractions & SDN 2 Research Examples 1.MPLS
2.Packet-Optical
Prototype Emulated WAN SAN FRANCISCO HOUSTON NEW YORK NOX
OpenFlow Protocol 25 GE links OC-48 links (2.5 Gbps)
Slide 26
Implementation of Control Architecture 26 NOX Interface:
OpenFlow Protocol Packet & Circuit Switches Converged Network
Unified Control Plane 1.Common Flow Abstraction 2. Common Map
Abstraction Application across packet and circuits
Slide 27
VOIP HTTP VOIP HTTP VIDEO Example Network Application Control
Function: Treat different kinds of traffic differently Function
Impl.: Use both packets and circuits, at the same time.
Traffic-typeDelay/JitterBandwidthRecovery VoIPLowest DelayLowMedium
VideoZero JitterHighHighest WebBest-effortMediumLowest
Slide 28
Video of a Demonstration of network application on Prototype 28
http://www.youtube.com/watch?v=Yq3yLPtOxpc&feature=relmfu
Why is it Simpler? 30 NOX Packet and Circuit Switches Converged
Network 2000 lines of code Unified Control Plane 1.Common Flow
Abstraction 2. Common Map Abstraction Application across packet and
circuits Interface: OpenFlow Protocol
Slide 31
Outline 2 Abstractions & SDN 2 Research Examples MPLS &
Packet-Optical 2 Burning Questions 1. Is Common Map practical? 2.
What about Scale?
Slide 32
32 1. Is Common Map practical? Extensibility NOX Packet and
Circuit Switches Converged Network Unified Control Plane 1.Common
Flow Abstraction 2. Common Map Abstraction Interface: OpenFlow
Protocol 1.Map Building & State- Distribution abstracted away
2. Full View
Slide 33
1. Is Common Map practical? Some well-known problems: Q:
Transport network operators dislike giving up (manual) control. Is
there a gradual adoption path? Q: IP and Transport networks will
not share information. How to build a common map? Answer to both:
Slicing Slice == Bandwidth + Switching
Slide 34
OpenFlow Protocol C CK P P P P Gradual Adoption Path CC Slicing
Plane Under Transport Service Provider (TSP) control ISP A Client
Controller OpenFlow Protocol ISP B Client Controller ISP C Client
Controller 34
Slide 35
Common Map PKT ETHETH ETHETH ETHETH ETHETH ETHETH ETHETH ETHETH
ETHETH ISP# 1s NetOS App PKT ETHETH ETHETH ETHETH ETHETH PKTPKT
ETHETH ETHETH SONETSONET SONETSONET TDMTDM ETHETH ETHETH ETHETH
ETHETH ETHETH ETHETH ETHETH ETHETH PKTPKT ETHETH ETHETH SONETSONET
SONETSONET TDMTDM ISP# 2s NetOS App PKT ETHETH ETHETH ETHETH ETHETH
PKTPKT ETHETH ETHETH SONETSONET SONETSONET TDMTDM Transport Service
Providers (TSP) virtualized network Internet Service Providers
(ISP# 1) OF enabled network with slice of TSPs network Internet
Service Providers (ISP# 2) OF enabled network with another slice of
TSPs network TSPs private line customer
Outline 2 Abstractions & SDN 2 Research Examples MPLS &
Packet-Optical 2 Burning Questions 1. Is Common Map practical? 2.
What about Scale?
Slide 40
Different Possibilities Control Plane Data Plane OpenFlow
Protocol Research and Prototyping Enterprise/DataCenter Networks
Carrier Networks
Slide 41
2. What about Scale? Limiting Resources Memory Compute
Consistency overhead DHTs DBs Resilience & Coordination
Mechanisms Zookeeper Dynamo Cassandra MongoDB Etc. Source: Martin
Casado, Nicira
Slide 42
2. What about Scale?
Slide 43
43 Why Does This Approach Scale? Per Packet Per Flow Per
Network Event No Consistency Eventual Consistency 10 6 10 8 /s 10 3
10 6 /s 10 1 10 3 /s Modification of Control Program Strong
Consistency0 - 10/s Source: Scott Shenker, Berkeley
Slide 44
Some Published Numbers 1 Onix instance 64 switches Cluster of 5
Onix instances Onix: A distributed control platform for large-
scale production networks. Teemu Koponen, et al. OSDI October 2010.
Latency < 300 us Throughput ~ 200,000 flows/sec Convergence
time: same or better than IGP
Slide 45
Outline 2 Abstractions & SDN 2 Research Examples MPLS &
Packet-Optical 2 Burning Questions 1. Is Common Map practical? 2.
What about Scale? 2 Trends 1.Research & Educational Networks
(RENs) 2.Commercial Networks
Slide 46
Trials (April 2010) 70+ trials/deployments in 13 countries
Slide 47
NSF GENI OpenFlow Deployment
Slide 48
Internet2 NDDI
Slide 49
OFELIA OpenFlow in Europe: Linking Infrastructure and
Applications
Slide 50
Outline 2 Abstractions & SDN 2 Research Examples MPLS &
Packet-Optical 2 Burning Questions 1. Is Common Map practical? 2.
What about Scale? 2 Trends 1.Research & Educational Networks
(RENs) 2.Commercial Networks
Slide 51
Home Network Carrier Network Data Center Network Cellular
Network Enterprise Network SDN Target Domains of Use
Slide 52
Open Networking Foundation (ONF) Mission Publish Open
Interfaces & Promote SDN
Slide 53
Windows (OS) Windows (OS) Windows (OS) Windows (OS) Linux Mac
OS Mac OS x86 (Computer) x86 (Computer) Windows (OS) Windows (OS)
App Linux Mac OS Mac OS Mac OS Mac OS Virtualization layer App
Controller 1 App Controller 2 Controller 2 Virtualization or
Slicing App OpenFlow Controller 1 NOX (Network OS) NOX (Network OS)
Controller 2 Controller 2 Network OS Computer IndustryNetwork
Industry Trend 53
Slide 54
Slide 55
Slide 56
Deutsche Telekom Verizon Google Microsoft Facebook Yahoo!
Comcast NTT Big Switch Networks Pronto Systems Midokura Nicira
Networks Plexxi Inc. Brocade Cisco Extreme Networks Force10
Networks Juniper Networks Huawei Technologies NEC Netgear Ciena
Ericsson Fujitsu Vello Systems Nokia Siemens Networks Dell HP IBM
Broadcom Intel Marvell Netronome Citrix VMware CompTIA Ixia
Mellanox Technologies Metaswitch Networks IP Infusion ONF
Membership Riverbed Technology Infoblox
Slide 57
Interop, May 2011 Demos: 16 companies demonstrated OpenFlow
Best of show: NEC OpenFlow switch
Slide 58
Summary 2 Abstractions 1.Flow Abstraction 2.Map Abstraction 2
Research Examples 1.MPLS 2.Packet-Optical 2 Burning Questions 1.Is
Common Map practical? Slicing 2.What about Scale? Distributed
Systems 2 Trends 1.Research & Educational Networks (RENs)
2.Commercial Networks ONF Software Defined Networks (SDN)
Slide 59
One Last Video http://www.youtube.com/watch?v=ie9EW4HmSwI