Read and Blue or Purple: Your Network, Your Way

Presenters : Gerard Phillips - Arista

Year : 2019

dlp_document_download : C U R A T E D B Y I P S H O W C A S E T H E AT E R AT N A B – A P R I L 8 -11 , 2 01 9 Red & Blue, or Purple Your network, your way Gerard Phillips, Systems Engineer, Arista Networks gp@arista.com , +44 7949 106098 What we ’ll cover • Architectural Overview – L2 vs L3 • Designing for Resilience • Architecture Options ‒ Monolithic ‒ Spine and Leaf - Hybrid ‒ Spine and Leaf – Purple • Conclusions 2 Architectural Overview – L2 • L2 networks typically deployed for audio installations ‒ Low bit rates ‒ Undersubscribed networks ‒ Control systems used L2 scoped discovery techniques ‒ MLAG provides scale, and spine resilience 3 Architectural Overview – L2 • But L2 does not work for Live Production, high bit rate multicast ‒ MLAG complex to configure for ASM Multicast (*,G) in a -7 environment ‒ Flows originated in remote switches are flooded towards the querier ‒ This potentially requires very large pipes! ‒ The failure domain is very large • You are also limited to 2 spines – potentially limiting scale 4 Querier Architectural Overview – L3 is the answer • This is the Datacenter architecture for scale and flexibility • PIM allows multicast to be routed • Failure domains are now able to be much smaller • Flooding towards the q uerier is no longer required ‒ Broadcast Controllers can be in charge of what transits any inter switch links 5 PIM PIM Unicast routing for L3 • Flexibility is one of the great benefits of the move to IP • To facilitate this flexibility, we need a solid unicast routing capability • This will under -pin any IGMP/PIM based multicast routing • BUT, can provide security, control, resilience and flexibility • Static routing can be used, but does scale … ‒ Manually provisioning routes is error prone and slow • BGP is the DC choice, scalable, fast convergence flexible, future proof ‒ But other dynamic routing protocols are available – OSPF, ISIS etc. 6 Designing for resilience • D etermined by how many failures your system should tolerate • -7 Hitless merge provides the capability to provide: ‒ RTP identical flows, on physically diverse NIC’s ‒ Physically diverse transport – optics, fibre, DAC, AOC etc ‒ Physically diverse IP fabric • You can survive the first failure, assuming you have a robust monitoring system that can provide quick, accurate, actionable info • You also have a path to planned maintenance, upgrade, addition of new services etc 7 Designing for resilience – the 2 nd failure • How do we survive the 2 nd failure? • Choose quality components – switches, NOS, optics, fibre etc. • Design in redundant PSU’s, Fans, Supervisors, Fabric Modules • Design in redundant Links between switches – N+1 or more • Ensure routing protocols, or SDN can, and will use these "spares" ‒ ECMP + LACP 8 Querier N+1 redundancy in the physical links Designing for resilience – smaller failure domains • Apply this physically as well as logically • Monolithic switches allow line -cards, fabric modules & supervisors to be replaced in service 9 Designing for resilience – smaller failure domains • Leaf and spine architectures allow you to manage smaller chunks: ‒ Route around failed components ‒ Route around devices under maintenance ‒ Influence multicast routing tables ‒ SDN 10 Monolithic • Simplicity. Hitless merge -7 resilience • SDN / flow orchestration is not necessary, IGMP can be used very successfully. • Monolithic chassis’ solutions can scale up to 16K 2 @ 3Gbe or 2304 hosts @ 25Gbe • Redundancy is provided by 2 (essentially) air -gapped switches, redundant fans and PSU’s, and optionally redundant switch supervisors 11 Monolithic - Expansion • While this architecture is simple, it does have a scale limit • Future expansion can build on a monolithic base, by using the monolithic switches as spine devices, adding SDN/orchestration, and hanging leaves from the "spine" • This path opens up higher levels of future expansion, but provides a simple start point 12 Spine and Leaf – Air - gapped Red and Blue • L3 topology for cloud scale - supports future expansion • Air -gapped provides flow security ( -7) • BGP routing for fast and reliable unicast convergence • PTP Boundary Clocks in L&S provides scale and accuracy 13 • A Flow Orchestrator or SDN system is needed • Simple -7 resilience still available • Simple leaf pair could be a starting point! Spine and Leaf – Air - gapped Red and Blue • L3 topology for cloud scale - supports future expansion • Air -gapped provides flow security ( -7) • BGP routing for fast and reliable unicast convergence • PTP Boundary Clocks in L&S provides scale and accuracy 14 • A Flow Orchestrator or SDN system is needed • Simple -7 resilience still available • Simple leaf pair could be a starting point! Spine and Leaf – Air - gapped Red and Blue (Hybrid) • Purple switches support single homed devices • Add as many "purple" switches as you need • This architecture requires an SDN controller, BUT the dedicated Red/Blue spines make it a simpler device 15 Spine and Leaf – Purple • L3 topology for cloud scale - supports future expansion • Flow security ( -7) provided logically , not physically • BGP routing for fast and reliable unicast convergence • BC PTP in both L&S provides scale and accuracy 16 • Any switch can support single homed devices • A Flow Orchestrator or SDN system is needed • Orchestrator is more complex than Red+Blue case • Simple leaf pair could be a starting point! Spine and Leaf – Purple • L3 topology for cloud scale - supports future expansion • Good starting point for a converged network later 17 Conclusions • Choose your architecture for your needs • Choose SDN or IGMP/PIM to solve your workflow challenges • Choose Cloud S cale IP infrastructure ‒ Provides many layers of resilience: ‒ Focus on Quality = Reliable SW/HW = low TCO + high uptimes ‒ Don’t let monitoring be an afterthought! ‒ L3 provides this reliability and resilience at scale • .... and limits the failure domain size ‒ Build in reliability, with redundancy • -7 Hitless merge • Redundant links (N+) • Resilient IP protocols – BGP, ECMP 18 C U R A T E D B Y I P S H O W C A S E T H E AT E R AT N A B – A P R I L 8 -11 , 2 01 9 Thank You Gerard Phillips, Systems Engineer, Arista Networks gp@arista.com , +44 7949 106098