Elastic vSphere: Now with More Stretch

Discusses the design considerations around building stretched VMware vSphere clusters

Transcript

1. Before we start • Get involved! • If you use Twitter, feel free to tweet about this session (use hashtag #tovmug) • I encourage you to take photos or videos of today’s session and share them online • This presentation will be made available online after the event
2. Elastic vSphere: Now with More Stretch Examining the design considerations for building stretched VMware vSphere clusters Scott Lowe, VCDX 39 vExpert, Author, Blogger, Geek http://blog.scottlowe.org / Twitter: @scott_lowe
3. Agenda • A quick review of stretched clusters • Stretched cluster considerations • Looking forward • Questions and answers
4. Stretched Cluster Review A quick review of vSphere stretched clusters
5. Stretched Cluster Review • Sometimes referred to as a “split cluster” • A stretched cluster is a cluster with ESX/ESXi hosts in different physical locations, usually different geographic sites • Stretched clusters are typically built as a way to create “active/active” data centers in order to: • Provide high availability across sites • Do dynamic workload balancing across sites
6. Stretched Cluster Review • HA/DRS are not required in a stretched cluster, although they are typically deployed • Many pros and cons of stretched clusters stem directly from the use of HA/DRS • Stretched clusters are not a requirement for long-distance vMotion
7. Stretched Cluster Considerations Examining some considerations on the use of stretched vSphere clusters
8. Stretched Cluster Considerations: vMotion • Stretched clusters are not a prerequisite for long-distance vMotion • However, vMotion does derive from benefit from stretched clusters: • Intra-cluster vMotions are highly parallelized • Inter-cluster vMotions are serial • Using a stretched cluster could offer benefits to disaster avoidance use cases
9. Stretched Cluster Considerations: Storage • Active/active data centers require read/write storage at both ends • Storage performance suffers otherwise • Storage vMotion required to fix the performance hit • There are a couple different ways of addressing this requirement; each approach has its benefits/drawbacks • Solutions are generally limited to synchronous (~100km) distances
10. Stretched Cluster Considerations: Storage • One solution is the stretched SAN solution • Stretching the SAN fabric between locations will usually involve multiple VSANs and inter-VSAN routing • Typically read/write in one location (read-only in the other) • Cross-connect topology allows hosts on both side to access read/write storage but introduces multipathing considerations • Implementations with only a single storage controller at each location create a SPoF (single point of failure)
11. Stretched Cluster Considerations: Storage • Another solution is distributed virtual storage • Distributes storage in a read/write fashion across multiple sites • Uses data locality algorithms to maximize cache benefits • Typically uses multiple controllers in a scale-out fashion • Needs a clustered file system for simultaneous host access • Must address “split brain” scenarios
12. Stretched Cluster Considerations: Storage
13. Stretched Cluster Considerations: Storage VM R/W Storage
14. Stretched Cluster Considerations: Storage VM R/W Storage
15. Stretched Cluster Considerations: Storage VM R/W Storage
16. Stretched Cluster Considerations: HA • vSphere HA can provide cross-site failover • However, vSphere HA is not currently “site aware” • You can’t control failover destination • You can’t designate or define things like: • Per-site failover capacity • Per-site failover hosts • Per-site admission controls
17. Stretched Cluster Considerations: HA • In vSphere 4.x, no more than 8 hosts in HA-enabled stretched clusters or you’ll run afoul of HA primary node limitations • Can only deploy 4 hosts or less per site per cluster • Ensures distribution of HA primary nodes • No supported method to increase the number of primary nodes or to specify HA primary nodes • vSphere 5 has a new HA architecture that eliminates this consideration
18. Stretched Cluster Considerations: DRS • You can balance load across sites using vSphere DRS • However, vSphere DRS (like HA) is not “site aware” • DRS host affinity rules can mimic “site awareness” • DRS host affinity rules are not dynamic • DRS host affinity rules create administrative overhead • DRS host affinity rules are defined and managed on a per- cluster basis
19. Stretched Cluster Considerations: Storage DRS • vSphere 5 introduces Storage DRS • Like vSphere DRS, Storage DRS is not “site aware” • Manually align datastore clusters with storage topology to avoid introducing unnecessary latency • User-defined storage capabilities and profile-driven storage could be used to help mimic site awareness • Watch for impact on storage replication/synchronization
20. Stretched Cluster Considerations: Networking • Stretched clusters add some networking complexity • More complex network configuration is required to provide Layer 2 adjacency (or its equivalent) • More complex networking required to address routing issues created by VM mobility • Technologies to address these concerns are new (OTV, LISP) and require networking expertise to configure and maintain • What about the virtual network design?
21. Router Router L2 Tunnel Stretched Cluster Considerations: Networking VM
22. Router Router L2 Tunnel Stretched Cluster Considerations: Networking VM
23. Router Router L2 Tunnel Stretched Cluster Considerations: Networking VM
24. Router Router L2 Tunnel Stretched Cluster Considerations: Networking VM
25. Stretched Cluster Considerations: Operations • Movement of VMs between sites (perhaps due to HA/DRS) could impact other areas: • Backups • Personnel/support • Disaster recovery/replication • Performance of multi-tier applications
26. Looking Forward… What will the future bring? Some speculation on the future of stretched vSphere clusters
27. The Future of Stretched Clusters • HA/DRS are simultaneously “greatest strength and greatest weakness” for stretched clusters • vSphere 5 improves the situation significantly • Stretched clusters would directly benefit from improvements to HA/DRS in these areas: • “Site awareness” • More scalable/dynamic DRS host affinity rule management (policy-based placement)
28. The Future of Stretched Clusters • Stretched clusters will benefit from further networking developments such as: • LISP (or equivalent) to decouple network routing from network identity • OTV, EoMPLS, or other equivalents to enable Layer 2 adjacency • Not yet clear how VXLAN will play in this space • In the longer-term future, the need for Layer 2 adjacency needs to be addressed and resolved
29. The Future of Stretched Clusters • Improvements in storage functionality will help stretched clusters • Active/active read-write storage at greater distances • Better handling of “split brain” scenarios • Better/more direct integration with replication for topologies with >2 sites (Sync-Sync-Async, for example)
30. Questions & Answers
31. Thank You