HCI Sizing Guide: Right-Sizing for Enterprise
We once helped a manufacturing client right-size their HCI cluster. Their original vendor quoted 10 nodes. After analyzing their actual workload, we deployed 6 nodes — saving $45,000 upfront and $12,000 per year in licensing. The catch? The original vendor sized based on theoretical maximums, not real usage.
HCI sizing is part science, part art. The science is the math. The art is knowing which assumptions to challenge. Here's our complete sizing methodology based on 50+ enterprise deployments.
Why Sizing Matters
Get HCI sizing wrong and you face one of two problems:
Over-provisioned (wasted money): You buy 10 nodes when 6 would have worked. That's $30,000-60,000 wasted on hardware you'll never fully use, plus higher annual licensing and power costs.
Under-provisioned (performance problems): You buy 4 nodes when you needed 6. VMs are slow, storage latency spikes, and your team spends months managing performance issues before finally upgrading.
The sweet spot is sizing for today's workload plus 30-40% growth over 3 years. Here's how to find it.
Step 1: Inventory Your Workloads
Before touching a sizing spreadsheet, collect this data for every VM:
CPU: Current utilization (average and peak). Don't use provisioned CPU — use actual usage.
RAM: Current usage and peak usage. Most VMs are over-provisioned on RAM.
Storage: Current disk size, IOPS requirements, and growth rate.
Network: Bandwidth requirements and latency sensitivity.
We use a simple spreadsheet for this. Export data from your hypervisor (VMware vCenter, Hyper-V, or Nutanix Prism) and consolidate it. Here's what a typical entry looks like:
VM: SQL Server | CPU: 4 cores, 35% avg utilization | RAM: 32GB, 28GB used | Storage: 500GB, 500 IOPS | Network: 2Gbps peak
The key insight: most VMs use 30-50% of their provisioned resources. Sizing based on provisioned capacity leads to 2x over-provisioning.
Step 2: Calculate Total Resource Needs
Add up all VM resource requirements. Here's a real example from a 200-VM enterprise deployment:
Total CPU needed: 180 cores (not 400 provisioned).
Total RAM needed: 1.2TB (not 2TB provisioned).
Total storage needed: 40TB (not 80TB provisioned).
Total IOPS needed: 25,000.
Now add overhead. HCI has inherent overhead for:
Hypervisor: ~5% of CPU.
Storage controller: ~10% of CPU and RAM.
Data protection (replication): ~20% storage overhead.
Cluster management: ~2-3% overhead.
Adjusted totals: 200 CPU cores, 1.4TB RAM, 50TB storage (with replication).
Step 3: Select Node Configuration
Now match your needs to available node configurations. Here are common enterprise configurations:
Entry Node: 16 cores, 128GB RAM, 2TB storage. Good for: small workloads, dev/test, branch offices.
Standard Node: 32 cores, 256GB RAM, 4TB storage. Good for: most enterprise workloads. This is our most popular configuration.
High-Performance Node: 64 cores, 512GB RAM, 8TB storage. Good for: database-heavy, VDI, or high-density environments.
For our 200-VM example (200 cores, 1.4TB RAM, 50TB storage), using standard 32-core nodes:
CPU: 200 cores / 32 cores per node = 6.25 → 7 nodes (round up).
RAM: 1.4TB / 256GB per node = 5.5 → 6 nodes.
Storage: 50TB / 4TB per node = 12.5 → 13 nodes (this is the bottleneck).
Storage drives the sizing. You need 13 nodes just for storage. But that gives you way more CPU and RAM than needed. Consider:
Adding disk shelves to each node for more storage without adding compute.
Using denser storage nodes (8TB or 12TB per node).
Offloading cold storage to a separate tier.
With 8TB storage nodes: 50TB / 8TB = 6.25 → 7 nodes. That balances nicely with CPU needs.
Step 4: Plan for Growth
Never size for today only. Add 30-40% buffer for 3 years of growth:
CPU: 200 cores × 1.35 = 270 cores → 9 nodes at 32 cores.
RAM: 1.4TB × 1.35 = 1.9TB → 8 nodes at 256GB.
Storage: 50TB × 1.4 = 70TB → 9 nodes at 8TB.
Final recommendation: 9 nodes. This gives you headroom for 3 years of growth without over-buying.
But wait — you need N+1 redundancy for HA. That means 10 nodes total (9 for workload + 1 for failover). In practice, most HCI platforms handle this automatically: 10 nodes with 10% reserved for HA gives you 9 nodes of usable capacity.
Step 5: Network Sizing
Network is often overlooked in HCI sizing. Don't make that mistake.
Inter-node bandwidth: Each node needs 10GbE minimum. For high-performance workloads, use 25GbE or bond multiple 10GbE links.
Storage network: HCI storage replication runs on the same network. Plan for 3-5x your storage I/O bandwidth on the network.
Management network: A separate 1GbE management network keeps administration traffic out of the data path.
For our 10-node example: 10GbE × 2 links per node (bonded) = 20Gbps per node. With 10 nodes, that's 200Gbps of aggregate inter-node bandwidth. More than enough for most enterprise workloads.
Sizing Tools and Calculators
Every major HCI vendor offers sizing tools:
Nutanix: Sizer tool (web-based, takes VM inventory and outputs node count).
VMware: vSAN Sizing Guide and ReadyNode Configurator.
Sangfor: Contact your sales rep for custom sizing (they do it manually).
Use the vendor tool as a starting point, then validate with your own analysis. We've seen vendor tools over-size by 20-30% to be "safe." Your own data is more accurate.
Best Practices
Measure first, buy second. Collect at least 30 days of utilization data before sizing.
Use actual usage, not provisioned capacity. Most VMs are over-provisioned by 50-100%.
Size storage separately. Storage is usually the sizing bottleneck, not CPU or RAM.
Add 30-40% growth buffer. 3 years is the standard planning horizon for HCI.
Plan for HA. Always add 1 node for redundancy. Don't count it as usable capacity.
Consider workload mix. Mixed workloads (VMs + databases + VDI) need more buffer than uniform workloads.
Conclusion
HCI sizing doesn't have to be guesswork. Inventory your workloads, calculate actual resource needs, add growth buffer, and match to node configurations. The math is straightforward — the discipline to collect accurate data is what separates good sizing from bad.
If you're facing an HCI sizing decision, start by exporting your current VM utilization data. That single spreadsheet will tell you more than any vendor sales pitch.
Want to go deeper? Explore [VMware alternatives](/en/vmware-alternative), [Run infrastructure services](/en/products/run), or [platform comparison](/en/compare).
FAQ
Q: How often should I re-evaluate HCI sizing?
A: Every 12 months. Run a capacity report, compare to your original sizing assumptions, and adjust. Most HCI platforms have built-in capacity dashboards.
Q: Can I mix different node sizes in the same cluster?
A: Yes, most HCI platforms support mixed-size nodes. But for optimal performance and management simplicity, use identical node configurations when possible.
Q: What if I undersize — can I add nodes later?
A: Yes, that's one of HCI's biggest advantages. Add nodes anytime. Storage and compute scale automatically. Just make sure you have budgetreserved for growth.
Q: How accurate are vendor sizing tools?
A: They're a good starting point but tend to over-size by 20-30%. Always validate with your own utilization data.
Sizing and Capacity Planning
Proper sizing is critical for HCI deployments. Start by inventorying your current workloads: CPU cores, memory per VM, storage per VM, and IOPS requirements. A general rule of thumb: each HCI node should run at 60-70% capacity to allow for growth and failover.
For a typical deployment of 50-100 VMs, we recommend starting with 4 nodes, each with: 2x 16-core CPUs, 256GB RAM, 4x 1.92TB NVMe SSDs, and 2x 25GbE NICs. This provides enough resources for most small-to-medium workloads with room to grow.
Migration Strategy from Traditional Infrastructure
Migrating from traditional SAN/NAS-based infrastructure to HCI requires careful planning. We recommend the following approach: First, identify non-critical workloads for initial migration (development, testing, staging environments). Second, use live migration tools (HCX for VMware, Xi Frame for Nutanix) to move VMs with zero downtime.
Third, validate performance on HCI before migrating production workloads. Monitor for 2-4 weeks to ensure IOPS, latency, and throughput meet requirements. Fourth, migrate production workloads in phases, starting with the least critical and progressing to mission-critical systems.
Disaster Recovery with HCI
HCI provides built-in high availability within a cluster, but you still need a disaster recovery plan for site-level failures. Options include: HCI-to-HCI replication between data centers (RPO as low as 5 minutes), cloud-based DR using HCI vendor cloud services, and hybrid DR with cloud object storage for backup.
We typically recommend a 3-2-1 backup strategy: 3 copies of data, on 2 different media types, with 1 copy offsite. With HCI, this translates to: local vSAN replication (copy 1), backup to secondary storage (copy 2), and cloud backup (copy 3).