1. Profile the workloads you're planning on virtualizing
   http://spiceworks-faqs.john-refactored.com/doku.php?id=virtualization:profiling
2. Make a guess about other workloads you'll be adding to the mix.
3. Apply 3-5 year growth factors to these workloads.
4. Configure hardware to meet the requirements
5. Shop the hardware among Tier 1 manufacturers
   Spiceworks has the RFQ tool for this

• Purchase from Tier 1 hardware providers with on-site repair contracts.
  Read erik-ptek's experience with Supermicro support.
  http://community.spiceworks.com/topic/post/1440240
• Use local storage until it's impossible to do so
• Realize that live workload migration between hosts is a higher cost feature. Don't assume you need it. Cost-justify it.
• If you need shared storage, investigate VSAs from VMware and HP before purchasing an external storage device.
  http://www.vmware.com/products/datacenter-virtualization/vsphere/vsphere-storage-appliance/overview.html
  http://h18006.www1.hp.com/products/storage/software/vsa/index.html
  
  HP has a two-node, 10 TB bundle for $6,000.
  http://h30094.www3.hp.com/product.aspx?mfg_partno=TA688SC
• Only buy external storage devices in a manner that matches eliminating single points of failure. That is, buying two hosts for failover but a single NAS/SAN has added configuration and management complexity without eliminating single points of failure. NAS/SAN needs to be purchased in pairs with automatic failover. Otherwise, what's the point?

For environment where there are identifiable “key” workloads that need to come back up and “less key” workloads that the environment can do without in the case of a hardware failure.

• Primary Host
  Sized to run all the workloads with headroom for future growth.
• Backup Host
  Sized to house replicas of all the workloads but only runs the key workloads in non-degraded mode. Possibly able to run all the workloads with performance compromises.
• Backup / Replication software to periodically create a replica of the workloads from the primary to the backup host.
• Requires hardware maintenance windows.

Create an environment where all the workloads can be run on a second host in the case of the first host failing.

• Primary Host
  Sized to run all the workloads with headroom for future growth.
• Backup Host
  Configured to have at least the same level of resources as the Primary.
• Backup / Replication software to periodically create a replica of the workloads from the primary to the backup host.
• Requires hardware maintenance windows, but if problems occur on the primary host during a maintenance window, the backup environment can be promoted to Primary.

When High Availability or vMotion/XenMotion/live-migration are business requirements.

• Primary Host
  Sized to run all the workloads with headroom for future growth.
• Backup Host
  Configured to have at least the same level of resources as the Primary.
• Shared Local Storage
  VMware VSAN, Starwind iSCSI SAN, HP Lefthand VSA, DRBD, etc. generally have a guest on each host that owns most of the local storage. The storage is maintained in a mirrored state and is then presented back to the hosts as shared storage.
  Best practice is to have separate “air gap” switching infrastucture.
• Since the hypervisor sees storage as shared, high availability features (auto guest reboot on backup host in case of primary host failure) and live migration are possible.
• You still need backup software.
• With live migration features, one generally doesn't need hardware maintenance windows.

When a virtualization environment's host count reaches 4 or more, the scalability of local shared storage becomes more complex, and centralizing shared storage in an appliance becomes more cost-effective.

• Host count is 4 or more
• Multiple host configurations are possible (but too many causes management overhead costs to rise)
• High Availability and migration features are available
• To avoid single points of failure, generally two appliances are needed with automatic failover
• Requires separate air-gapped switching infrastructure with redundant paths to the appliances
• You still need backup software
• You need a separate backup target