§ · the case for ceph

Storage that doesn't
rent itself back to you.

Ceph is the open-source, software-defined storage system that powers CERN, the Australian BoM, and a long list of national labs and hyperscalers you've never heard of. One cluster, one pool of disks — object (S3-compatible), block, and file, with no per-protocol licensing. Replication, self-healing, and rebalancing are built in. You add commodity servers, you get more capacity. You lose a server, the cluster keeps serving.

01

You own the bytes.

No egress fees. No tiering tax. No vendor pulling the rug at renewal. The disks live in your rack — including a lot of jurisdictions where storing customer data anywhere else is a compliance fail.

02

One cluster, three protocols.

S3 (RGW), block (RBD), and POSIX file (CephFS) on the same pool of disks. Stop paying for an object store, a SAN, and an NFS appliance separately.

03

Scales linearly.

Need more? Rack a server, claim its disks, walk away. Ceph rebalances. No forklift, no migration window, no “please contact your account manager.”

§ · do the math

What does 1 PB cost you, really?

Slide to your target usable capacity. We'll size a real Ceph cluster on reference Supermicro hardware and compare 3-year TCO against the major cloud object stores at public list pricing.

Usable capacity 63 TB
on-prem

Ceph on Supermicro

one-time capex · then power, cooling, spares

Upfront (year 0) $52K
Then per year $3K
Supermicro SSG-640P-E1CR36L 3 × server · 36-bay 4U, 2× Xeon, 256 GB ECC, 2× 25 GbE
$36,000
22 TB enterprise SATA 9 × drive · Seagate Exos / WD Ultrastar class
$3,150
Top-of-rack switching redundant 25/100 GbE leaf pair · scales every 16 nodes
$8,000
Rack, PDU, cabling, install $1,500 / server allocation
$4,500
Hardware capex (one-time)
$51,650
Power, cooling, spares 5% of capex / year — typical colo & ops
$2,583/yr
Raw 189 TB
Replication
Built capacity 66 TB
hyperscaler

Cloud object storage

recurring · public list pricing · storage only

Upfront (year 0) $0
AWS S3 per year $18K
AWS S3 Standard $0.023 / GB-month · $1K/mo
$17,805/yr
Google Cloud Storage Standard $0.020 / GB-month · $1K/mo
$15,483/yr
Azure Blob Hot (LRS) $0.0184 / GB-month · $1K/mo
$14,244/yr
Backblaze B2 $0.006 / GB-month · $387/mo
$4,645/yr

Storage charges only. Egress, PUT/GET requests, lifecycle transitions, and Glacier retrieval fees not included — they only widen the gap.

§ · cumulative spend, year by year

When does Ceph pay for itself?

Break-even vs AWS 3 yr 5 mo
YearCeph (cumulative)AWS S3 (cumulative)GCP (cumulative)Azure (cumulative)Δ vs AWS
upfront$51,650$0$0$0+$51,650
Year 1$54,233$17,805$15,483$14,244+$36,427
Year 2$56,815$35,611$30,966$28,488+$21,204
Year 3$59,398$53,416$46,449$42,733+$5,982
Year 4$61,980$71,221$61,932$56,977−$9,241
Year 5$64,563$89,027$77,414$71,221−$24,464
vs AWS S3 Standard 3 yr 5 mo
vs GCP Standard 4 years
vs Azure Blob Hot 4 yr 5 mo

At 63 TB usable, the 3-node Ceph minimum means a higher upfront — but break-even with AWS lands at 3 yr 5 mo, and every disk you add after that scales without touching the chassis. Ceph wins for any workload you'll keep longer than the break-even horizon.

Assumptions and how the numbers are built
  • Replication: 3× replicated pool — Ceph's safe default. Erasure-coded pools (e.g. EC 4+2) reduce raw overhead to 1.5×, cutting hardware costs by another ~50% with a small CPU and write-latency tradeoff.
  • Reference server: Supermicro SSG-640P-E1CR36L (36 × 3.5" top-load bays, dual Xeon Scalable, 256 GB DDR4 ECC, dual 25 GbE, mirrored NVMe boot). Listed at $12,000 in a typical OEM channel build.
  • Drives: 22 TB SATA enterprise (Seagate Exos X22 / WD Ultrastar DC HC580 class) at $350 each, list. Bulk pricing typically lands lower.
  • Networking: Redundant 25/100 GbE leaf pair (e.g. Mellanox SN2100 / Arista 7050 class) at $8,000 per pair. Scales one pair per ~16 nodes.
  • Minimum cluster: 3 nodes, for monitor quorum and failure-domain isolation. Below ~50 TB usable, the 3-node floor dominates the capex.
  • Power & cooling: 5% of capex per year. Real colo + electricity varies widely by region — adjust to your facility.
  • Cloud pricing: US-region public list, hot tier, single-region redundancy where applicable. Q1 2026.
  • Not included on either side: staff cost, software licensing (Ceph itself is Apache 2.0; Sentinel is its own line item), data transfer / egress, request charges, backup-of-backup.

Ready to do this in production?

Sentinel is the operator console that makes a Ceph cluster feel like an appliance — provisioning, lifecycle, monitoring, and every storage protocol from one browser tab.