UPDATED: Virtualizing MS SQL Server on vSphere 8.0 Architecture Best Practices Guide

You have met the new, improved VMware vSphere 8.0. You have met the Windows Server 2022. You have, indeed, been geting chummy with Microsoft SQL Server 2022. You have, right?

No? You haven’t? Doh! We sincerely apologize for our tardiness. We were busy putting finishing touches to the masterpiece we were creating, not only to ensure that we cover all the bases, but to cover them in a way that the end product doesn’t get mistaken for anything other than the comprehensive and authoritative reference architecture it has always been – since the invention of Pipes.

In addition to updating existing sections to reference the latest versions of all Products in focus, this updated version of the Architecting Microsoft SQL Server on VMware vSphere also includes discussion of new or expanded capabilities, features and practices such as:

  • Configuring CPU Hot Plug in vSphere 8.0 underwent a major uplift, given the changes introduced in both vSphere 8.0 and Windows Server 2022. We described this in details and also re-visited the “CPU Hot Plug and ‘Phantom NUMA Node'” issue and demonstrated how it’s no longer an issue with these releases.
  • We devoted some time to celebrating (sorry, introducing) the fact that, in vSphere 8.0, enabling CPU Hot-Add for a Windows VM no longer disables vNUMA. We then spent some time describing SQL Server Automatic Soft-NUMA and vNUMA compatibity and behavior.
  • Configuring virtual NUMA for “Wide” Microsoft SQL Server VMs? What is a “Memory-Wide VM”, and how does vSphere 8.0 handle its vNUMA topology? We provided extensive descriptions, with multiple configuration scenarios to cover most use cases.
  • VMware introduced virtual Hyperthreading (vHT) in vSphere 8.0 as an enhancement to the “Latency Sensitivity” setting which has long existed in vSphere. This changes most of what you knew about improving performance for vSphere-hosted HT-aware Business-critical application workloads.
  • Now, there is no technical need to use LSI Logic SAS Controllers for your Windows VMs anymore. Go all-PVSCSI for the maximum storage IO throughput possible.
  • We addressed the now-prevalent “Cluster-on-Die” (aka Sub-NUMA Clustering) hardware feature and its implications on large (“Wide”) VMs on vSphere. Although we did not prescribe a recommendation, we hope that our prescriptive guidance provides enough details to enable the reader to make an informed decision for their specific configurations, situations and use cases.
  • We briefly described how, with the new vSphere vMotion Notifications for Latency Sensitive Applications feature in vSphere 8.0, application owners and administrators are now better able to more finely control and schedule vMotion events, significantly improving SQL Server workloads’ availability and resilience. vMotion notification also complements native SQL Server HA capabilities, especially for planned outages and workloads relocation.
  • Clustered VMDK as a replacement for Raw Device Mapping (RDM) requirement for shared-disk clustering? We covered that, too.
  • Should you consider moving from vSAN Original Storage Architecture (OSA) to vSAN Express Storage Architecture (ESA) for your MS SQL Server workloads? Check out the Guide.

We hope that you find this Updated Document useful, and we encourage you to please provide feedback through the appropriate channels (or your account reps). Your input helps us make these documents better and more fit for purpose.

Thank you.