The economics of hyper-convergence

Intermediate⏱ ~17 minceph

What you'll learn

Quantify the hardware saving
Identify the costs that offset it
Compare the two architectures on total cost
Recognise where the economics change

Prerequisites

None — start here.

Verified against Ceph Tentacle 20.2.x · Ceph Squid 19.2.x (supported previous) · cephadm matches the verified Ceph release · podman 4.x · csi-rbd and csi-cephfs current · RBD / CephFS / RGW current (matches Ceph release) · Linux kernel 5.15+ (5.10 minimum) · Ubuntu 24.04 LTS (Ceph host baseline) · Debian 12 (Bookworm) (Ceph host baseline) · Rocky Linux / RHEL / AlmaLinux 9.x (Ceph host baseline) · Proxmox VE 9.x (cross-course integration) · Kubernetes 1.31+ (cross-course integration) · 2026-08-18

Why this matters in production

Hyper-convergence is chosen on cost, and the comparison is usually made on hardware alone. The other terms are real and change the answer at scale.

The hardware saving

Dedicated, 3 Proxmox + 3 Ceph nodes:
  3 × compute-optimised host  (high CPU, high RAM, minimal storage)
  3 × storage-optimised host  (moderate CPU, moderate RAM, many drives)
  = 6 chassis, 6 sets of network ports, 6 rack units of power

Hyper-converged, 3 nodes:
  3 × balanced host  (high CPU, high RAM, many drives)
  = 3 chassis, 3 sets of network ports, 3 rack units of power

The saving is roughly half the chassis, half the network ports, and half the rack space and power — at small scale a substantial fraction of the total.

Where the saving disappears

Cost	Effect
Over-provisioning for contention	must size for compute peak plus storage
Reduced usable capacity	1/N host absorption on a small cluster
Longer maintenance windows	both layers coordinated
Higher operational complexity	one system, two concerns
Wasted capacity on scaling	adding for one need adds the other

3-node hyper-converged, size=3:
  host failure absorption = 1/3 = 33% of usable capacity reserved
  and with size=3 on 3 hosts, a host failure means no third copy at all

The three-node case is the sharpest: size=3 with failure_domain=host on three hosts means a host failure leaves PGs undersized until it returns, regardless of free space.

Comparing honestly

Total cost of ownership over 3 years:
  hardware (chassis, drives, network)
  power and cooling
  rack space
  operational time (maintenance, incidents, coordination)
  capacity efficiency (usable fraction of raw)
  risk cost (correlated failure exposure)

Scale	Typical winner
3 nodes	hyper-converged, clearly
4–8 nodes	hyper-converged, usually
8–12 nodes	roughly even
12+ nodes	dedicated, increasingly

Where the economics change

# the ratio that drives it
ceph osd df | awk 'NR>1 {n++} END {print n, "OSDs"}'
# against VM count and CPU utilisation

Signal	Indicates
VM density growing faster than storage need	compute-bound; dedicated storage frees the ratio
Storage growing faster than VM need	storage-bound; same argument reversed
Contention incidents recurring	the coupling is costing more than the hardware saved
Maintenance windows lengthening	coordination cost is growing

Quiz

Knowledge check · 4 questions

Q1. What happens when a host fails in a three-node hyper-converged cluster with size=3 and failure_domain=host?

PGs stay undersized until the host returns, because CRUSH has nowhere to place a third copy Recovery redistributes the data across the remaining two hosts once the down-out interval elapses The pool drops to size=2 automatically until the host returns, then restores the third copy Writes are blocked immediately, since a size=3 pool cannot acknowledge with only two copies
Q2. The hardware saving from hyper-convergence is largest at exactly the scale where a host failure cannot be recovered from.

True False
Q3. Evaluate architecture for a new deployment.

A new deployment needs compute for roughly 200 VMs and about 80 TB of usable storage. The team is evaluating hyper-converged against dedicated on cost.
Q4. Why does hyper-convergence waste capacity when the compute-to-storage ratio is skewed?

Passing score: 75%. Answers are checked in this browser.

Production discipline

Understand that a three-node hyper-converged cluster with size=3 and failure_domain=host cannot restore redundancy after a host failure — the PGs stay undersized until it returns. Size compute and storage requirements independently before choosing; a skewed ratio is what makes the coupled architecture expensive.

Cross-course references

Kubernetes: co-locating storage and workloads couples their scaling identically
Linux: any shared-resource design trades flexibility for hardware efficiency