1 / 31

Demystifying Fabrics When, Where and Why…

Demystifying Fabrics When, Where and Why…. Paul Unbehagen Chief Architect Avaya Networking. @ punbehagen. It’s On!. A friendly competition among presenters Help @ punbehagen win Easy to remember name @ randy_cross ? Too forgettable Tweet this session with # AvayaATF and #SDN

coye
Download Presentation

Demystifying Fabrics When, Where and Why…

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Demystifying FabricsWhen, Where and Why… Paul Unbehagen Chief Architect Avaya Networking @punbehagen

  2. It’s On! • A friendly competition among presenters • Help @punbehagen win • Easy to remember name • @randy_cross? Too forgettable • Tweet this session with #AvayaATF and #SDN • You could win a iPad Mini… • If you don’t have a twitter account… • Get one, you can delete after the show • Help me WIN by coming to my other sessions (and Tweeting!) • Demystifying Fabrics (#Fabrics) • Deployment Option for Avaya VENA DToR (#dToR)

  3. The (Data Center) Network Fabric..?

  4. The Agile Network..!

  5. Not all Fabrics are the same Replace Spanning Tree Protocol OSPF PIM MPLS BGP LDP Replace Spanning Tree Protocol ?

  6. Which Fabric Technology is the Answer..? That all depends on how you qualify the question… Avaya Extensions Avaya VENA Fabric Connect • Abstraction • Service-based virtualization • Orchestration-ready • Aspirational functionality • But it requires: • BGP • LDP • RSVP-TE • Draft-Rosen • VPLS • Baseline redundancy • Root Bridge –dependent • Not shortest path • Single logical Switch / fault domain • 100m distance limitation • VLAN-based virtualization • Root Bridge –dependent • Large flooding domain • VLAN-based virtualization • Layer 3 Awareness • Unicast & Multicast support • Application-driven extensibility Application Awareness IETF MPLS L3 Multicast Virtualization L3 Unicast Virtualization IEEE SPB – Multi-Vendor L2 Multi-Site Virtualization IETF TRILL Cisco FabricPath Brocade VCS Juniper QFabric L2 Single-Site Virtualization L2 Multi-Pathing STP L2 Loop-free Topology

  7. Spanning Tree 802.1Q MPLS/BGP/LDP/RSVP-TE/Draft-Rosen/VPLS TRILL/FabricPath SPB IETF (draft Unbehagen) SPB IEEE Technology Compared L2 loopfree Topology L2 Multipathing L 2 Virtualization L 3 Virtualization Other virtualization: • - IPv6 Virtualization • Application based Virtualization • … Unicast Multicast Multiple/ hosted DC Single DC Root Bridge Dependency Not shortest path Root Bridge Dependency Large Flooding Domain Vlan based Service based

  8. Host X Host Y MAC H MAC D MAC Z MAC C MAC N MAC B MAC F MAC I Rbridge A TRILL Nickname TA Rbridge E TRILL Nickname TE Router B Router C Router D SPB D SPB B SPB A SPB C SPB E Host X Host Y Forwarding Comparison Outer Layer 2 header is replaced at each hop with an appropriate Layer 2 header for the next hop and a hop count is decreased TRILL X Y MPLS Eth TA | TE N | F X | Y FCS’’ TA | TE Z | B FCS X | Y Payload Inner-Eth TRILL Outer-Eth Outer-Eth Inner-Eth Payload TRILL TA | TE H | I X | Y FCS’’’ C | D TA | TE X | Y FCS’ Payload TRILL Outer-Eth Inner-Eth Payload Inner-Eth TRILL Outer-Eth SPB Route Lookup Y X FCS X | Y A | E Eth Payload SPB is much simpler, lower cost, OAM-transparent solution

  9. Loop Handling… SPB TRILL & FabricPath MAC-B MAC-B MAC-A MAC-A Ingress SPB ForwardingDatabase 2/11 MAC-A 2/12 MAC-B TTL 2/11 2/12 2/11 2/12 • TTL allow loop and discard it after value reach 0 • Give up on the problem, dimensioning the crater.. • SPB’s RPFC (Reverse Path Forwarding Check) does not allow loops • Prevents Loops before they begin

  10. Provisioning New Services..? • Causes: • Based on VLANs • Touch-points everywhere • ..? • Impact: • Error prone • Slows time-to-service • Constrains agility • Virtualization unfriendly • ..?

  11. Subnets Spanning Data Centers..?

  12. Migrating Applications & Virtual Machines..?

  13. Start With Carrier-Grade Foundations Instantaneous Recovery

  14. Encompass Data Center, Campus Core & Edge End-to-End

  15. Need To Add New Services..? Edge-Only Provisioning

  16. Need To Add Multicast..? Receiver Receiver Multicast- for-Free Sender

  17. Need To Add Greater Separation..? Integrated Routing & VRF

  18. Need To Add More Capacity..? Real-Time Maintenance

  19. SPB’s Forwarding Model All traffic is forwarded in the most optimal path based on SPF calculations Replication done at optimal point based on shortest path algorithm Packets only traverse SPF links, no out of order packets, only nodes that are a member of the same service receive packets Multicast traffic originates at Edge-1. One multicast packet sent to Core-1. Edge-8 Edge-9 Edge-1 Edge-6 Core-1 Core-2 Edge-2 Edge-5 Edge-3 Unicast and Multicast always follow the same path No intervention needed, no root bridges, just turn it on VLAN evolves into a Service with simple end point provisioning Edge-4 Common VLAN service

  20. TRILL’s Forwarding Model Problemunicast and multicast can take different paths In Trill model, unicast and multicast take different paths Replication NOTdone at optimal point, but to the SPF to the root bridge Multicast traffic originates at Edge-1. One multicast packet sent to a root bridge. Edge-8 Edge-9 Root Bridge Edge-1 Edge-6 Core-1 Core-2 Unicast can take a different path, risking reordering packets Edge-2 Edge-3 Edge-5 Edge-4 Common VLAN service

  21. TRILL’s Forwarding Model Problemthe solution is worse then the original problem… Make every edge a ROOT Bridge! To get best multicast model they have to make every edge a root bridge Lots of manual config of where root bridges exist Causing every edge to receive a packet whether it has a service or not Packets sent to non-service participating nodes, just to be dropped Multicast traffic originates at Edge-1. One multicast packet sent to Core-1. Edge-8 Edge-9 Root Bridge Edge-1 Edge-6 Core-1 Core-2 Edge-2 Edge-3 Edge-5 Edge-4 Common VLAN service

  22. Avaya’s Approach

  23. Game-Changing Functionality Fast Three pillars of value to Fabric Connect • Provision at the “edge” • One Configuration Command • Optimized Link State Protocol • Fast to Converge, heal,& add, delete, move services Flexible • Extend services anywhere seamlessly • True service virtualization with ease • L2, L3, Multicast, VRFs… Secure • As much service isolation as needed • Carrier type virtualization, zero complexity • Network Invisibility to users

  24. An Innovative Approach to a Growing Problem ERS 8800 Fabric Connect Core VSP 9000 VSP 9000 North-South/Core-ToR Interconnects VSP 7000 Distributed Data Center SDSN SDSN Distributed Top-of-Rack SDSN

  25. Flexible Network Services Trill and FabricPath can only do L2 • Mapping of a Layer 2 VLAN into a Virtual Service Network delivering seamless Layer 2 extensions • Layer 2 Virtual Service Network Virtual Service Network Virtual Service Network Virtual Service Network • Mapping of a Layer 3 VRF into a Virtual Service Network delivering seamless Layer 3 extensions • Layer 3 Virtual Service Network Virtual Service Network SPB enables all service types • Enhancing 802.1aq by offering a policy-based Layer 3 internetworking capability of multiple Virtual Service Networks • Inter-VSN Routing • Native IP routing across the Virtual Service Fabric without the need for Virtual Service Networks or any additional IGP • IP Shortcuts VLAN VLAN

  26. Technology Evolution Completely unique solution to the real DC problem • 3-D design • Optimized for actual traffic flow • True, extensible virtualization for Networking, Compute, and Storage Fabric Connect

  27. Layer 2 (E-LAN) Virtualized Service Layer 3 Virtualized Unicast Service Layer 3 Virtualized Multicast Service Layer 3 Multicast Service Service Oriented Networking Network Service Layer Customer Service Layer Layer 3 Unicast Service VXLAN Service E-Line Service ADN/SDN Service E-Tree Service “ACME“ Control Plane RFC 6329 IS-IS / 802.1aq (SPBm) 802.1ah (MACinMAC) (2-16 BVLANs) Forwarding Plane SPB Layer Connectivity Layer E-LINE/VPWS (PBB - or MPLS based) E-LAN/VPLS (two end-points) Dark Fiber CWDM/ DWDM Multiple Service Providers Infras-tructure Layer Physical Infrastructure SONET, SDH, Ethernet, etc… The Benefit of SPB Network Service Layer is independent from infrastructure provider

  28. Integrated Multicast value Powerful Integrated IP Multicast support for an alternative to PIM • Key Characteristics: • Standard IGMP at the Access • Autonomic within the Fabric • No need for PIM or DVMRP complexity • L2 or L3 Virtual Services Networks • Use Cases: • IP Video Surveillance • IPTV • VMware VXLAN integration • Availability: • ERS 8800 • both Edge & Core • VSP 9000 • Core; Edge to be added in 3.4 Multicast SenderGroup 239.0.0.10 Receiver IGMP Join 239.0.0.10 Receiver ISIS ISIS IGMP Join 239.0.0.10 ISIS ISIS ISIS ISIS IPMC ISIS ISIS ISIS IGMP Join 239.0.0.10 ISIS ISIS IGMP Join 239.0.0.10 Receiver Receiver

  29. Summary • Standards • SPB: IEEE 802.1aq Shortest Path Bridging • 802.1aq supports two modes, SPB VID (SPBV) and SPB MAC (SPBM) where the ERS 8600/8800 supports SPBM • SPBV uses Q-in-Q IEEE 802.1ad encapsulation • SPBM uses mac-in-mac IEEE 802.1ah (Supported by Avaya and others) • IEEE protocols that have already been deployed in carriers and enterprises around the world • FabricPath • No standards, completely proprietary from Cisco • TRILL • TRILL is a IETF standard reinventingIEEE protocols that have already been deployed in carriers and enterprises around the world • KEY NOTES • SPB has successfully demonstrated multi-vendor interoperability • http://ieee802.org/1/files/public/docs2011/aq-ashwood-smith-spbm-3rd-interop-0718-v01.pdf • FabricPath is totally proprietary – how does a Cisco customer migrate to TRILL? • With TRILL, no two vendors implementation is the same; no multi-vendor interoperability!!

  30. @punbehagen

More Related