Posts Tagged ‘Design’

Top 6 VMware vSphere Design Questions

April 27, 2011 Leave a comment

Should I use a distributed vSwitch or a standard vSwitch?

  • Do you need to delegate network configuration or the advanced functionality a dvSwitch offers?
  • Trade convenience for additional considerations
  • Network configuration now dependent upon vCenter Server
  • Affects running vCenter Server as a VM
  • Requires Enterprise Plus licensing
  • A “hybrid” approach utilizing both vSwitches, where possible, provides the best of both worlds

Should I run vCenter Server as a virtual machine?

  • Both options are fully supported by VMware
  • Virtual has advantages (can leverage HA, for example)
  • Physical has advantages (no dependencies on the infrastructure it manages)
  • Virtual introduces new considerations:

–      Need vCenter for dvSwitch control plane

–      What if vCenter is VM and runs across dvSwitch?

–      Creates circular dependency

–      Operational concerns with DRS, EVC, VUM

Should I use blades or rack mount servers?

  • From a compute perspective, it’s a wash
  • The impact falls primarily in high availability

–      Must consider HA cluster size and cluster members per chassis

–      Can’t use redundant cards in blades in many instances (no redundant NICs or HBAs)

  • Newer blades offer as much connectivity as many rack mount servers (12 NICs, dual HBAs)
  • More exotic connectivity (InfiniBand, FCoE, PCIe extenders) not as widespread

Should I choose VMware ESX or VMware ESXi?

  • A common but not long-lived question
  • vSphere 4.1 is the last version to contain VMware ESX; all future versions will use only ESXi
  • So, perhaps a better question is, “How can I transition to ESXi?”
  • One step is to familiarize yourself with the vSphere CLI and/or the vSphere Management Assistant
  • If you’re a CLI junkie, get used to “vicfg-” instead of “esxcfg-”

Should I put all my hosts into a single large cluster

  • There is no one right answer!
  • Are you using blades?

–      Keep <5 cluster members per chassis

–      Must scale number of chassis to scale cluster size

  •  With vSphere 4.1 and VAAI, SCSI reservation conflicts are not a gating factor
  • Clusters are not vMotion boundaries, only DRS/HA/FT organizational unit

Should I lump all my VMs together in one big LUN?

  • LUN layout should be driven more by I/O profile than capacity or number of VMs
  • VAAI hardware-assisted locking eliminates SCSI reservation conflicts
  • There are potential performance benefits to multiple LUNs (multiple queues per LUN)
  • Less management overhead with fewer LUNs
  • The key is proper storage design to accommodate I/O requirements
Categories: VMware Tags: , ,