Real world FabricPath Deployment at IBM Data Centers Santiago Freitas CCIE#18776 (R&S / SP) Consulting Systems Engineer Cisco-IBM Global Team safreita@cisco.com Lasse Leegaard IT Architect AT&T lasse@intl.att.com Real world FabricPath deployment at IBM Data © 2012 Cisco and/or its affiliates. All rights reserved. Centers Cisco Cisco Confidential Confidential 1 What? IBM has achieved tangible benefits by migrating its infrastructure from Catalyst 6500 to a Nexus 2000, 5000, 7000 Architecture. IBM has adopted FabricPath on the Nexus 5K and 7K and MPLS L3VPN on the Nexus 7K. The solution was extensively tested at Cisco ECATS. FabricPath was a key differentiator when competing with Juniper. We learned a lot from this deployment. Real world FabricPath deployment at IBM Data © 2012 Cisco and/or its affiliates. All rights reserved. Centers Cisco Confidential 2 Session Objectives At the end of the session, you should be able to: Articulate to your customers the Business Benefits that IBM has achieved by migrating to a Nexus 2K/5K/7K Architecture. Explain the reasons why they adopted FabricPath. Understand the Tests performed to validate the solution before deployment. Understand IBM’s future direction and how they plan to get there. Real world FabricPath deployment at IBM Data © 2012 Cisco and/or its affiliates. All rights reserved. Centers Cisco Confidential 3 IBM Nordic Strategic Outsourcing One of the company’s largest Integrated Market Teams (IMT) globally IBM SO provides outsourcing services that offer management of applications and other IT components in either an onsite or hosted arrangement. Eight Data Centers located in Denmark, Sweden and Finland. Serve around 200 customers ‒ Some have dedicated infrastructures ‒ Over 100 served by a shared, multitenant infrastructure Real world FabricPath deployment at IBM Data © 2012 Cisco and/or its affiliates. All rights reserved. Centers Cisco Confidential 4 Overview of the Network Infrastructure Overall network structure Customers Telcos Internet AS25384 Si Dedicated gateways hold 1 telco Shared gateways holds multiple telcos One shared set hold direct customer connections on 1G and internet access MPLS gateway: 6500, 7200 or 7300 Si MPLS route reflectors VLAN + IP MPLS LDP link Si Si MPLS gateways MPLS P routers MPLS layer MPLS P router: 6500 or 7600 L2 trunk VPLS PE routers Si Si Si Server access block Si MPLS PE routers / Aggregation switches Layer 2 trunks 10G or Nx1G Si Si Service switches FWSM/NAM/ACE Si Si Access switch: 6500 Service switch: 6500 MPLS PE router / Aggregation switch: 6500 VPLS PE router: 7600+ES20 Limit of 20 access switches is based on un-oversubscribed port density in the Core routers Up to 20 access switches Real world FabricPath deployment at IBM Data © 2012 Cisco and/or its affiliates. All rights reserved. Centers Cisco Confidential 5 Overview of the Network Infrastructure ~150 Cisco 6500/7600 ~50 Cisco 7200/7300 ~ 3000 VLANs and 290 virtual firewalls ~26000 Ethernet ports Real world FabricPath deployment at IBM Data © 2012 Cisco and/or its affiliates. All rights reserved. Centers Cisco Confidential 6 One of the Access Blocks Reached EoL A portion of the shared infrastructure was approaching end of life Cisco and AT&T performed EoL analysis. Factual discussion: Vital to demonstrate the need for a full network refresh. - 22 Access Switches - 4080 access ports 1G or 2x1G uplinks - 2 pair of FWSM in Service Switches Real world FabricPath deployment at IBM Data © 2012 Cisco and/or its affiliates. All rights reserved. Centers Cisco Confidential 7 Hardware Refresh Options Replace only of the parts that reached End of Life Technology Refresh using Catalyst 6500 Technology Refresh using Nexus 2K/5K/7K Does not solve the High Risks and Technology Limitations Does not solve some of High Risks and Technology Limitations. Limited evolution 10G uplinks and Resolves single point of failure issues Solves all of High Risks and Technology Limitations Hidden cost of line cards replacement within 3 years $2.440.430,00 (now) + $2.460.000,00 (2014) = $4.900.430,00 Total Rack Space – 390 RU Total Rack Space – 154 RU Total Power – 156 KW Total Power – 60 KW Total cost - $5.378.635,00 Total cost - $2.639.300,00 Real world FabricPath deployment at IBM Data © 2012 Cisco and/or its affiliates. All rights reserved. Centers Cisco Confidential 8 Business Benefits of the Nexus-based Solution Why IBM chose to deploy Nexus and FabricPath Significant OPEX savings when compared with the existing infrastructure: ‒ Reduced the power consumption by 61% ‒ Reduced the rack space used by network switches by 60% ‒ Reduced the number of managed devices in the network by 38.5% (from 26 to 16) Easier way to scale, supports more access blocks on the same Core devices, therefore less expensive per customer port Reduction in the time to onboard and configure the network for new customers CAPEX savings – Next Generation DC based on Cisco Nexus and FabricPath was 46% cheaper than building similar architecture using Catalyst 6500 Real world FabricPath deployment at IBM Data © 2012 Cisco and/or its affiliates. All rights reserved. Centers Cisco Confidential 9 Juniper!!! Yes – we did consider doing it differently Like-for-like (EX8200/4500/4200 + MX routing + 6500/FWSM firewall) Qfabric (Qfabric switching + MX routing + 6500/FWSM firewall) No FCoE capable hardware 10G server density not impressive FCoE development is beginning to catch up However, Nexus has more/longer field exposure than Juniper kit in this area. Organizational inertia and training would have to be overcome Real world FabricPath deployment at IBM Data © 2012 Cisco and/or its affiliates. All rights reserved. Centers Cisco Confidential 10 What IBM actually deployed? 2x Nexus 7010 MPLS Backbone ‒ M1/F1 combination ‒ MPLS L3 VPN PE 12x Nexus 5548UP FabricPath ‒ Across 3 DCs 70x Nexus 2200 ‒ 3360 access ports 2x 6500 Service chassis for FWSM modules Real world FabricPath deployment at IBM Data © 2012 Cisco and/or its affiliates. All rights reserved. Centers Cisco Confidential 11 FabricPath Flexibility The Network Can Evolve With No Disruption Need more edge ports? Need more bandwidth? L3 → Add more leaf switches → Add more links and spines L3 FabricPath FabricPath Real world FabricPath deployment at IBM Data © 2012 Cisco and/or its affiliates. All rights reserved. Centers Cisco Confidential Why IBM adopted FabricPath? vPC and traditional STP topologies were considered Better utilization of links Increased Agility ‒ New PODs and/or links for more capacity can be added non-disruptively ‒ Any VLAN anywhere Simplicity of Configuration ‒ Much simpler to implement and configure than vPC Very fast convergence - sub-second in most cases Need to route over the Fabric ‒ Layer 3 over FabricPath Real world FabricPath deployment at IBM Data © 2012 Cisco and/or its affiliates. All rights reserved. Centers Cisco Confidential 13 FabricPath enablement Was that really it? install feature-set fabricpath vpc domain 11 feature-set fabricpath role priority 100 vlan 3865 peer-keepalive destination 10.1.20.46 source 10.1.20.45 mode fabricpath spanning-tree mst configuration name IBMMST02 revision 10 peer-gateway auto-recovery fabricpath switch-id 1000 instance 1 vlan 1-2048 instance 2 vlan 204 interface Ethernet1/5 switchport mode fabricpath fabricpath domain default spf-interval 50 50 50 lsp-gen-interval 50 50 50 root-priority 255 / 254 (N7K) fabricpath switch-id 1 Real world FabricPath deployment at IBM Data © 2012 Cisco and/or its affiliates. All rights reserved. Centers Cisco Confidential 14 MPLS L3 VPN on Nexus 7000 Works together with the rest of the infrastructure Nexus 7010 as the MPLS L3VPN PE. Customer VLANs mapped into VRF/VPN in the Aggregation Layer. Remote Sites are 6500, 7600, 7300 and 7200, working well with the rest of the infrastructure. Nexus 7010 Advantage over Juniper, extra layer required. Real world FabricPath deployment at IBM Data © 2012 Cisco and/or its affiliates. All rights reserved. Centers Cisco Confidential 15 Real world FabricPath deployment at IBM Data © 2012 Cisco and/or its affiliates. All rights reserved. Centers Cisco Confidential Migration plan How to get from here to there (or from there to here depending on your point of view) MPLS P VPLS PE MPLS PE + Aggregation L3 L3 FabricPath VLANs VLANs FW/LB service + Access Real world FabricPath deployment at IBM Data © 2012 Cisco and/or its affiliates. All rights reserved. Centers Cisco Confidential 17 ECATS End of Test Report Cisco Enhanced Customer Aligned Testing Services - http://ecats 36 Major Tests Areas Detailed Results DDTS/Bugs Found and workarounds Technical Notes Convergence Summary Table HW and SW utilized Lessons Learned Configuration Files See Additional Resources Slides for link to it Real world FabricPath deployment at IBM Data © 2012 Cisco and/or its affiliates. All rights reserved. Centers Cisco Confidential 18 Migration plan How to get from here to there (or from there to here depending on your point of view) MPLS P VPLS PE MPLS PE + Aggregation L3 L3 FabricPath VLANs VLANs FW/LB service + Access Real world FabricPath deployment at IBM Data © 2012 Cisco and/or its affiliates. All rights reserved. Centers Cisco Confidential 19 ECATS End of Test Report Cisco Enhanced Customer Aligned Testing Services - http://ecats 36 Major Tests Areas Detailed Results DDTS/Bugs Found and workarounds Technical Notes Convergence Summary Table HW and SW utilized Lessons Learned Configuration Files See Additional Resources Slides for link to it Real world FabricPath deployment at IBM Data © 2012 Cisco and/or its affiliates. All rights reserved. Centers Cisco Confidential 20 ECATS testing experience Cisco Enhanced Customer Aligned Testing Services - http://ecats Vital on the success of this deployment. Gives us experience before having used it Test overlap with rollout Reduction of risk of introducing new technology Real world FabricPath deployment at IBM Data © 2012 Cisco and/or its affiliates. All rights reserved. Centers Cisco Confidential 21 IXIA 4/9 IXIA 4/10 9/1 Testing Topology 9/2 Access5 6500 10.53.234.172 L2 Access Layer 9/47 Remote Site 9/48 L2 1G 8/1 8/1 Core3 7600 10.53.234.170 L2/L3 Aggregation MPLS PE BGP / ISIS / MP-BGP / LDP 9/3 9/1 M1 ports 1/1 Core1 Nexus 7010 10.53.234.166 10.53.234.167 L2/L3 Aggregation MPLS PE BGP / ISIS /MP-BGP / LDP 9/4 MP-BGP Peering between PEs 9/2 L3 - 10G MPLS/LDP enabled links ISIS 9/2 3/9 Po 10 4/9 4/9 F1 ports3/31 VPC+ peer link 4/1 3/2 - ISIS / MP-BGP / LDP ‒ Access Layer Cat6500 (Layer 2) M1 ports 1/1 2/1 3/9 3/1 - L2/L3 Aggregation 9/1 L3 - 10G MPLS/LDP enabled links ISIS 2/1 3/31 ‒ Agg/MPLS PEs (7600) Core4 7600 10.53.234.171 L2/L3 Aggregation MPLS PE BGP / ISIS / MP-BGP / LDP 9/3 Po 10 9/4 4/2 3/1 4/1 F1 ports 3/2 Core2 Nexus 7010 10.53.234.168 10.53.234.169 L2/L3 Aggregation MPLS PE BGP / ISIS / MP-BGP / LDP ISIS and MPLS in the core 4/2 Site Under Test F1 ports LACP Gi 0/1 Gi 0/2 ‒ Nexus 7010 as Agg/MPLS PE (L2/L3) Switch 3 10.53.234.148 FaE 0/3 IXIA 4/16 L2-10Gbs FabricPath Enabled Links 1/1 Access1 Nexus 5548UP 10.53.234.162 2/13 1/2 1/3 1/4 2/14 Po 10 1/3 FEX 100 Po20 1 100 /1/10 100/1/1 100 100/1/2 /1/48 IXIA 4/1 Gi 0/1 L3 BGP / OSPF connections BGP FaE 0/0/0 CE1-2851 Access3 1/1 Nexus 5548UP 10.53.234.164 2/13 Access2 Nexus 5548UP 2/14 10.53.234.163 1 100/1/46 100 /1/47 1/2 FEX 100 Po20 2 Nexus 2248 IXIA 4/11 1/1 1/4 2/13 IXIA 4/2 IXIA 4/3 Gi 0/1 2 100 /1/10 1 100 /1/47 2/13 IXIA 4/5 FaE 0/3 1 Gi 0/1 BGP CE2-2821 IXIA 4/7 OSPF FaE 0/1/8 100 /1/48 ‒ Nexus 5548UP/Nexus 2248 as Access Switch 2 ‒ FabricPath ‒ Servers attached with vPC+ IXIA 4/8 FaE 0/2 10.53.234.147 FaE 0/3 CE4-2851 10.53.234.174 10.53.234.169 IXIA 4/14 100 /1/10 FaE 0/1 LACP ‒ vPC+ at the Core for Active/Active HSRP 2 Nexus 2248 100 /1/47 IXIA 4/6 FaE 0/0/0 CE3-2851 10.53.234.173 10.53.234.167 2/14 Access4 Nexus 5548UP 10.53.234.165 FEX 100 Po20 100/1/1 100 100/1/2 /1/48 Gi 0/1 L3 BGP / OSPF connections 1/2 100 /1/10 100/1/46 FaE 0/2 Switch 1 10.53.234.146 OSPF FaE 0/1/8 1/1 1/3 Po 10 1/4 2 Nexus 2248 IXIA 4/13 100 /1/48 IXIA 4/4 FaE 0/1 LACP 1/3 1/4 2/14 FEX 100 Po20 Nexus 2248 100 /1/47 1/2 ‒ OSPF/BGP over FP IXIA 4/15 Note any IP addresses shown are for management purposes only. Title: Description: IBM Voyager Author: mraines Proposed topology for ECATS Nexus 7K/5K 7600 testing Updated: 13/03/2012 Filename: Minimal-Testing-topology-FabricPathv19.vsd Real world FabricPath deployment at IBM Data © 2012 Cisco and/or its affiliates. All rights reserved. Centers Cisco Confidential 22 Testing Topology and Scale Numbers For Your Reference Hardware and Software Versions and Scale Numbers Access Layer ‒ Nexus 5548UP – NX-OS 5.1(3)N1(1a) ‒ FEX Nexus 2248 Core ‒ Nexus 7010 – NX-OS 5.2(3a) ‒ 2x M1 8x 10GE (N7K-M108X2-12L) ‒ 2x F1 32x 1/10GE (N7K-F132XP-15) Remote Site PE ‒ 7609 with RSP-720 – IOS 15.1(1)S 300 VLANs 300 SVIs and 300 HSRP 200 VRFs / MPLS L3 VPN 3000 MAC addresses injected Real world FabricPath deployment at IBM Data © 2012 Cisco and/or its affiliates. All rights reserved. Centers Cisco Confidential IMIX Ethernet Traffic ‒ 4Gbps within Nexus Access Block (East 23 Convergence Times Sub-second on FabricPath link failures Failover Test Result Convergence Summary Layer 3 Link Failure on Core towards Remote site – 64 ms / 30 ms on Recovery M1 Line Card Failure on Core - 950 ms (North-South) / 75 ms on Recovery Fabric Path Link Failures (multiple tests) – 117 ms / 241 ms on Recovery F1 Line Card failure on Core - 1380 ms / 319 ms on Recovery Core Node Failure (power off N7010) - 2584 ms / 2703 ms on Recovery Access Node Failure - 316.52 ms for vPC+ attached servers / 181 ms on Recovery Real world FabricPath deployment at IBM Data © 2012 Cisco and/or its affiliates. All rights reserved. Centers Cisco Confidential 24 Dynamic Routing Protocol and FabricPath You can run OSPF and BGP over FabricPath, you can’t over vPC The OSPF CE routers CE-3 and CE-4 were configured with “ip ospf priority 0” interface configuration so they don’t participate in DR/BDR election process FULL OSPF neighborships are formed with both Core1 and Core2 CE3-2851-RK18#sh ip ospf neighbor Neighbor ID Pri State Dead Time Address Interface 10.10.101.5 1 FULL/BDR 00:00:36 10.10.101.5 GigabitEthernet0/1 10.10.101.7 1 FULL/DR 00:00:33 10.10.101.7 GigabitEthernet0/1 10.10.101.8 0 2WAY/DROTHER 00:00:30 10.10.101.8 GigabitEthernet0/1 CE3-2851-RK18# Traffic still forwarded even when crossing peer-link FabricPath doesn’t have same limitations as vPC Real world FabricPath deployment at IBM Data © 2012 Cisco and/or its affiliates. All rights reserved. Centers Cisco Confidential 25 Technical Lessons Learned It would be a session on its own… No Show Stopper DDTS ‒ One cosmetic, one catastrophic but with an easy workaround (already fixed) and one Unreproducible. Several Technical Lessons Learned on the areas of: ‒ Peer-Link Failure and vPC+ attached devices ‒ MAC Learning with vPC+ domain ‒ Multidestination tree and vPC+ ‒ MAC Learning on N7K with M1/F1 for L2 Traffic Details on the hidden slides and on Additional Resources page Real world FabricPath deployment at IBM Data © 2012 Cisco and/or its affiliates. All rights reserved. Centers Cisco Confidential 26 Further developments Where do we see the rest of the infrastructure go? Real world FabricPath deployment at IBM Data © 2012 Cisco and/or its affiliates. All rights reserved. Centers Cisco Confidential 27 Evolution Plan MPLS P Routers Layer 3 / MPLS MPLS PE/ Agg Switches Layer 2 IPv4 Services Switches (FWSM/ACE/NAM) SAN A MPLS PE/ Agg Switches Layer 2 IPv4 Services Switches Up to 20 Access Switches (FWSM/ACE/NAM) SAN B SAN A VPLS PE Layer 2 IPv4 Services Switches Up to 20 Access Switches (FWSM/ACE/NAM) SAN B SAN A VPLS PE Services Switches Up to 20 Access Switches (FWSM/ACE/NAM) SAN B Layer 2 IPv4 Layer 2 IPv4 SAN A Services Switches Up to 20 Access Switches (FWSM/ACE/NAM) SAN B SAN A Up to 20 Access Switches SAN B 28 Management Orchestration Provisioning Automation Evolution Plan MPLS P Routers Dynamic Infrastructure Layer 3 / MPLS 2^12 = 4096 VLANs… 2^24 = 16777216 Segment IDs MPLS PE/ Agg Switches Layer 2 IPv4 IPv4/IPv6 Services Switches (FWSM/ACE/NAM) MPLS PE/ Agg Switches Layer Layer22 IPv4IPv6 IPv4/IPv6 IPv4/ Services Switches Up to 20 Access Switches (FWSM/ACE/NAM) SAN B SAN A Layer 2 IPv4 Services Switches Up to 20 Access Switches (FWSM/ACE/NAM) Storage 1 2 FC/FCoE/NAS 3 4 SAN A VPLS PE SAN A SAN B VPLS PE Layer 2 IPv4 Layer 2 IPv4 Services Switches Up to 20 Access Switches (FWSM/ACE/NAM) Services Switches Up to 20 Access Switches (FWSM/ACE/NAM) Up to 20 Access Switches Storage 1 2 FC/FCoE/NAS 3 4 A SAN B SAN B SAN A SAN B SAN A SAN B 29 Key Takeaways The Key Takeaways of this presentation were: IBM has achieved OPEX and CAPEX savings by migrating to a Nexus 2K/5K/7K Architecture in their Data Centers. IBM has adopted FabricPath and is very happy with its Flexibility, Easy to Implement and Use and Convergence Time. FabricPath was extensively tested and validated at Cisco ECATS. FabricPath and MPLS on N7K were differentiators against Juniper. You can reuse the lessons learned and additional resources available from this deployment to position FabricPath to your customers. Real world FabricPath deployment at IBM Data © 2012 Cisco and/or its affiliates. All rights reserved. Centers Cisco Confidential 30 Additional Resources You can find the following additional information on the link below ‒ Customer Requirements and Business Case for Catalyst 6500 -> Nexus and FabricPath ‒ Joint Technical Plan of Record (test requirements) ‒ Detailed Test Plan ‒ Complete end of Test Report (including detailed test results and configurations) ‒ Lessons Learned Presentation ‒ INTERNAL Case Study of IBM Nordic Adoption of Nexus and FabricPath ‒ EXTERNAL version of the Case Study http://bock-bock.cisco.com/wiki/User:Safreita:FabricPath_Testing Real world FabricPath deployment at IBM Data © 2012 Cisco and/or its affiliates. All rights reserved. Centers Cisco Confidential 31 Presentation_ID © 2012 Cisco and/or its affiliates. All rights reserved. Cisco Public