Demystifying SDN Saurav Das sauravdasalumni stanford edu ATT

  • Slides: 35
Download presentation
Demystifying SDN Saurav Das sauravdas@alumni. stanford. edu AT&T Talk 3/27/14 1

Demystifying SDN Saurav Das sauravdas@alumni. stanford. edu AT&T Talk 3/27/14 1

A number of slides are courtesy of www. sdnacademy. com 2

A number of slides are courtesy of www. sdnacademy. com 2

Outline • What is SDN? (or how was it originally intended) • Different Shades

Outline • What is SDN? (or how was it originally intended) • Different Shades of SDN • SDN Myths and Misconceptions

What is SDN? (B) State-Distribution Abstraction: Global View Network OS running on distributed controllers

What is SDN? (B) State-Distribution Abstraction: Global View Network OS running on distributed controllers Configuration; Control over Forwarding; Monitoring (A) Forwarding Abstraction: Match-Action Tables © 2013 SDN Academy, LLC™. All Rights Reserved. 4

What is an Abstraction? Data – plane abstractions Where are the control-plane abstractions? ©

What is an Abstraction? Data – plane abstractions Where are the control-plane abstractions? © 2013 SDN Academy, LLC™. All Rights Reserved. 5

Where are the Control-Plane Abstractions? • Networking is about getting stuff from here to

Where are the Control-Plane Abstractions? • Networking is about getting stuff from here to there … • Is OSPF a topology abstraction? • Is OSPF a routing abstraction? Developing a Control • Can I use it to route IPv 6 traffic? Plane Abstraction • well you need v 2 for IPv 4 and v 3 for IPv 6 Requires • Can I use it to route multicast traffic? • no you need MOSPF or PIM Developing a Data Plane • Can I use it in Ethernet networks? (Forwarding) Abstraction • no you need TRILL (with IS-IS) for that Distributed Network Functions State Distribution Mechanisms PE Label Distrib ution LDP Switch Operating System © 2013 SDN Academy, LLC™. All Rights Reserved. Route Advert VPNIPv 4 Route Advert TE Label Distrib ution IGPRoute Advert, Link. State I-BGP + RR MPBGP RSVPTE OSPF v 2 E-BGP learned 6

Towards a Forwarding Abstraction 7

Towards a Forwarding Abstraction 7

(A) Forwarding Abstraction A way to define forwarding behavior that is not layer or

(A) Forwarding Abstraction A way to define forwarding behavior that is not layer or vendor specific 8

Interfaces are the Realizations of Abstractions 9

Interfaces are the Realizations of Abstractions 9

(B) State Distribution Abstraction Control program should not have to handle distributedstate details Proposed

(B) State Distribution Abstraction Control program should not have to handle distributedstate details Proposed abstraction: global network view Abstracted away by Network Operating System State Collection Dissemination & Synchronization Application Isolation 10

So What is SDN? Control Abstraction: Global View App App northbound interface Controller southbound

So What is SDN? Control Abstraction: Global View App App northbound interface Controller southbound Interface eg. Open. Flow Switch Forwarding Abstraction: Match-Action Tables Switch Interfaces are the Realizations of Abstractions 11

What is SDN? – A Simpler View SDN has two defining characteristics. First, it

What is SDN? – A Simpler View SDN has two defining characteristics. First, it physically separates the control plane (which decides how to handle the traffic) from the data plane (which forwards traffic according to decisions that the control plane makes). Second, SDN consolidates the control plane, so that a single instance controls multiple data- plane elements -- The Road To SDN, http: //queue. acm. org/detail. cfm? id=2560327 © 2013 SDN Academy, LLC™. All Rights Reserved. 12

Benefits of SDN 1. Simpler Control with Greater Flexibility • Networks work because we

Benefits of SDN 1. Simpler Control with Greater Flexibility • Networks work because we can master complexity, but what we should be doing is extracting simplicity, with the right abstractions 2. Programmability: Speed of Innovation, Ease of Service Insertion & Faster Time to Market • Does not involve changing/creating a fully distributed protocol 3. Lower Total Cost of Ownership (TCO) • Lower Opex – easier to manage, troubleshoot, emulate, automate, optimize • Lower Capex – replacing proprietary hardware, pay for what you need and no more. © 2013 SDN Academy, LLC™. All Rights Reserved. 13

[1] ‘Pure’ SDN / ‘True’ SDN? Separation D <---> C Simplicity (B) State-Distribution Abstraction:

[1] ‘Pure’ SDN / ‘True’ SDN? Separation D <---> C Simplicity (B) State-Distribution Abstraction: Global View Network OS running on distributed controllers Programmability Configuration; Control over Forwarding; Monitoring Lower Capex Lower Opex (A) Forwarding Abstraction: Match-Action Tables © 2013 SDN Academy, LLC™. All Rights Reserved. 14

Different Shades of SDN • ‘Pure’ SDN or ‘True’ SDN • Orchestration/ Automation/ Gluework

Different Shades of SDN • ‘Pure’ SDN or ‘True’ SDN • Orchestration/ Automation/ Gluework • Traditional Networking + SDN ‘Hook’ • Overlay Networks for DC Network Virtualization • Traditional Networking + White-Box + Orchestration • Decoupled Traditional Networking + Global View • Open Source Networking © 2013 SDN Academy, LLC™. All Rights Reserved. 15

[2] Orchestration/Automation/Gluework Provides Automation Separation D <---> C “Controller” Simplicity By Orchestrating Across EMS

[2] Orchestration/Automation/Gluework Provides Automation Separation D <---> C “Controller” Simplicity By Orchestrating Across EMS Vendor Y Vendor Z Programmability CLI Vendor X Automation Modified NMS, Open. Stack, cli-scripts automate Configuration & provisioning © 2013 SDN Academy, LLC™. All Rights Reserved. Lower Capex Programmability Changing Behavior (if-then-else) Lower Opex 16

[3] Traditional Networking + ‘SDN Hook’ Legacy Preserving with Claimed Advantages A. Customer realizes

[3] Traditional Networking + ‘SDN Hook’ Legacy Preserving with Claimed Advantages A. Customer realizes benefits of SDN on same infrastructure B. Allows vendor to build SDN on existing products ‘Controller’ CLI Replacement Instead of CLI expect scripts, write python programs Under traditional networking control PBR overide L 2 L 3 Tunnel © 2013 SDN Academy, LLC™. All Rights Reserved. ECMP ACL 17

[3] Traditional Networking + ‘SDN Hook’ Legacy Preserving with Claimed Advantages A. Customer realizes

[3] Traditional Networking + ‘SDN Hook’ Legacy Preserving with Claimed Advantages A. Customer realizes benefits of SDN on same infrastructure B. Allows vendor to build SDN on existing products Point Solutions PCE Exampe: PCE based PCEP BGP-LS © 2013 SDN Academy, LLC™. All Rights Reserved. RSVP, LDP OSPF v 2, OSPF v 3, ISIS MP-BGP I-BGP + RR MOSPF, PIM 18

[3] Traditional Networking + ‘SDN Hook’ Legacy Preserving with Claimed Advantages A. Customer realizes

[3] Traditional Networking + ‘SDN Hook’ Legacy Preserving with Claimed Advantages A. Customer realizes benefits of SDN on same infrastructure B. Allows vendor to build SDN on existing products Hybrid Solutions / ‘Ships-in-thenight’ ‘Controller’ © 2013 SDN Academy, LLC™. All Rights Reserved. Some VLANs have SDN control 19

[3] Traditional Networking + ‘SDN Hook’ Separation D <---> C Legacy Preserving with Claimed

[3] Traditional Networking + ‘SDN Hook’ Separation D <---> C Legacy Preserving with Claimed Advantages A. Customer realizes benefits of SDN on same infrastructure B. Allows vendor to build SDN on existing products CLI Replacement ‘Controller’ Point Solutions Instead of CLI expect scripts, write python programs Simplicity Programmability Eg. PCE based Lower Capex Hybrid Solutions / ‘Ships-in-thenight’ ‘Controller’ © 2013 SDN Academy, LLC™. All Rights Reserved. Some VLANs have SDN control Lower Opex 20

[4] Overlay Networking Open. Stack, Cloud. Stack, VMWare R 2 R 1 Orchestrator Separation

[4] Overlay Networking Open. Stack, Cloud. Stack, VMWare R 2 R 1 Orchestrator Separation D <---> C B 2 B 1 B 3 R 3 Simplicity ‘Controller’ R 1 B 3 Underlying L 2/L 3 fabric v. Switch, v. Router Context: Data center network virtualization for public/private cloud; presented from the viewpoint of the cloud provider (not the tenant)! © 2013 SDN Academy, LLC™. All Rights Reserved. Programmability Lower Capex Lower Opex R 2 R 3 B 2 21

[5] Traditional Networking + White Box Separation D <---> C Simplicity Control Plane (Software)

[5] Traditional Networking + White Box Separation D <---> C Simplicity Control Plane (Software) Data Path (Hardware) White -box Programmability Lower Capex Lower Opex Merchant Silicon © 2013 SDN Academy, LLC™. All Rights Reserved. 22

[5. 5] Traditional Networking + White Box + Orchestration Separation D <---> C Simplicity

[5. 5] Traditional Networking + White Box + Orchestration Separation D <---> C Simplicity Control Plane (Software) Data Path (Hardware) White -box Programmability Lower Capex Lower Opex Merchant Silicon © 2013 SDN Academy, LLC™. All Rights Reserved. 23

[6] Decoupled Traditional Networking Separation D <---> C Simplicity Traditional networking control plane 1.

[6] Decoupled Traditional Networking Separation D <---> C Simplicity Traditional networking control plane 1. Software stacks 2. Merchant Silicon Usually proprietary, could be OF, See IETF For. CES © 2013 SDN Academy, LLC™. All Rights Reserved. Programmability Lower Capex Lower Opex 24

[6. 5] Decoupled Traditional Networking + Global View Separation D <---> C ‘Controller’ Simplicity

[6. 5] Decoupled Traditional Networking + Global View Separation D <---> C ‘Controller’ Simplicity Traditional networking control plane 1. Software stacks 2. Merchant Silicon Usually proprietary, could be OF, See IETF For. CES © 2013 SDN Academy, LLC™. All Rights Reserved. Programmability Lower Capex Lower Opex 25

[7] Open Source Networking Separation D <---> C Has Open Source ever worked in

[7] Open Source Networking Separation D <---> C Has Open Source ever worked in networking? What are the best know open source networking projects… • Quagga • OVS … not used in production without modification What about controllers? : • No open source controller used in production • No open source networking project today is complete solution • Open source is part of the solution Simplicity Programmability Lower Capex Lower Opex © 2013 SDN Academy, LLC™. All Rights Reserved. 26

Or c /A hest uto ra ma tion tio n Tra SD diti N

Or c /A hest uto ra ma tion tio n Tra SD diti N ‘ on Ho al n ok et w. ’ Ov + e for rla DC y N ne etw t v or irt k s. Tra W diti hit on e B al ox net w. Tra + d W iti hit on e B al ox net + O w. De rch + c. Tra oup dit led ion al Ne De tw. n/ coup w + G led lob Tra Op al d. Vie e So n w ur ce Tru SD e N Different Shades of SDN Separate D <---> C Simplicity Programmability Lower Capex Lower Opex © 2013 SDN Academy, LLC™. All Rights Reserved. 27

Myths & Misconceptions • SDN is Network Virtualization • SDN == NFV • SDN

Myths & Misconceptions • SDN is Network Virtualization • SDN == NFV • SDN is about centralized control • Open. Flow is not mature © 2013 SDN Academy, LLC™. All Rights Reserved. 28

Open. Flow has evolved towards production readiness. state 1. 0 Q 4 ‘ 09

Open. Flow has evolved towards production readiness. state 1. 0 Q 4 ‘ 09 1. 1 Q 1 ‘ 11 flows ports + Group Tables + Multiple Tables/Pipelines: 1. 2 1. 3 Q 4 ‘ 13 + optical ports + synchronized tables © 2013 SDN Academy, LLC™. All Rights Reserved. single message queue w/optional barriers + forward 1 -in-n (ECMP) + match Qin. Q, MPLS, SCTP + match virtual ports + per-flow metering + tunnel-id Q 2 ‘ 12 1. 4 forward {0, 1, n} match Eth, VLAN, IP, L 4 + IPv 6 + multiple controllers Q 4 ‘ 11 Interface msg behavior + extensible match + extensible actions + multiple channels (auxiliary connections) + bundle messages

Myths & Misconceptions • SDN is Network Virtualization • SDN == NFV • SDN

Myths & Misconceptions • SDN is Network Virtualization • SDN == NFV • SDN is about centralized control • Open. Flow is not mature • Open. Flow does not work with current hardware • Open. Flow does not scale • Do we really need another protocol? • I can’t just throw away my existing network … © 2013 SDN Academy, LLC™. All Rights Reserved. 30

31 Source: Ed Crabbe, Google

31 Source: Ed Crabbe, Google

32 Source: Ed Crabbe, Google

32 Source: Ed Crabbe, Google

33 Source: Guru Parulkar

33 Source: Guru Parulkar

Source: Guru Parulkar 34

Source: Guru Parulkar 34

Q&A

Q&A