Implementing VMware vSphere: from design to production

Design: start with a question, not hardware

The most common mistake is to buy servers and only then think about what will run on them. The order is reversed. First you need answers to these:

• How many virtual machines, and what CPU, RAM and disk demands?
• What availability does the business need — how much downtime can it afford?
• What's the three-year growth plan?
• What budget and licensing are realistic?

Hosts (ESXi)

Two is the minimum, three is the sensible standard. Principles:

• Identical host configuration — simplifies migration and planning.
• Plenty of RAM — it's usually the first resource to run out.
• Redundant power and network cards.
• A dedicated vCenter enabling vMotion, HA and DRS.

Storage

For vMotion and HA to work, all hosts must access shared storage.

Shared array (SAN/NAS)

A central array over iSCSI or NFS. A proven model, but the array itself must be redundant.

vSAN

Storage from disks in the hosts, pooled into a resilient whole. No separate array, but it needs a careful design and a fast network.

Plan for headroom. Storage at 90% behaves unpredictably and drops performance.

Network

Split traffic into separate networks so they don't compete:

• Management (vCenter, ESXi).
• vMotion.
• Storage (iSCSI/NFS or vSAN) — ideally on separate cards.
• Production VM traffic.

Each type should have redundancy — two cards and two switches where budget allows.

HA and DRS

• HA — on host failure, automatically restarts its machines elsewhere. Downtime from hours to minutes.
• DRS — continuously balances load across hosts.

Both only make sense when sized to N+1.

Backup and monitoring

Virtualization is not a backup. HA protects against hardware failure, not deletion or ransomware.

• Regular VM backups outside the cluster.
• Verify backups — an untested backup is not a backup.
• Monitoring and email alerts for faults.

What to avoid

• A cluster with no headroom (N+1).
• Local storage instead of shared — vMotion and HA won't work.
• Neglected network and power redundancy.
• Assuming HA replaces a backup.