Mr.PlanB Logo
    Virtualization Storage Ledger 2026
    Ledger
    Virtualization Storage Ledger 2026

    Ceph: Virtualization Storage Ledger 2026 Scorecard

    The open-source engine inside IBM Storage Ceph, OpenShift Data Foundation, croit and Proxmox VE: block, file and object at any scale, with no vendor of its own.

    Edition October 16, 20268 min read

    Where Ceph put its effort, 2016 to 2026

    Each square is one capability area in one year. The brighter it is, the more the area gained in the Virtualization Storage Ledger that year; 1, 2 and 3 mark that year's biggest pushes.

    ’16’17’18’19’20’21’22’23’24’25’26
    Data services core11
    Resilience and HA11
    Performance and protocols212
    Hypervisor integration2
    Scale and topology
    Data protection and DR12
    Security23
    Operations and automation131
    Hardware and deployment freedom
    Kubernetes and containers11
    no change+0.5+1+1.5 or moreScores out of 5 per area.

    Biggest pushes overall: Kubernetes and containers (+4), Performance and protocols (+2), Operations and automation (+2). Already strong in 2015: Hardware and deployment freedom, Scale and topology, Resilience and HA.

    Top focus each year: 2016 Data protection and DR; 2017 Data services core; 2018 Operations and automation; 2019 Kubernetes and containers; 2020 Operations and automation; 2021 Resilience and HA; 2022 Kubernetes and containers; 2023 Performance and protocols; 2024 Resilience and HA; 2025 Data services core. 1 of 11 years show no gain in any area.

    Part of the Virtualization Storage Ledger (edition 2026-10-16). How the scores work: methodology.

    Snapshot

    • Type: engine (open source; packaged commercially by IBM, Red Hat, croit, Proxmox and others)
    • Owner: No single owner. Ceph Foundation, a directed fund under the Linux Foundation since 2018-11-12; technical governance by the Ceph Steering Committee; the largest employer of core developers moved from Red Hat to IBM on 2023-01-01
    • Licence: LGPL-2.1 or LGPL-3 for the main code with component exceptions; free, no editions, no licence change found since 2023
    • Latest release: Tentacle 20.2.4 (2026-08-19); Squid 19.2.7 (2026-10-12); Tentacle 20.2.0 shipped 2025-11-18
    • Support lifecycle: One named release a year; each stable series gets backports for about two years; point releases every four to six weeks in principle; upgrades tested from the last two stable releases, no downgrade. Reef ended 2026-03-20, Squid estimated 2026-10-31, Tentacle estimated 2027-06-01
    • Next milestone: Umbrella (21.x): shown as in development at Ceph Days London (June 2026), no GA date published
    • Hypervisors: QEMU/KVM hosts natively through librbd and the kernel RBD client (Proxmox VE, OpenStack, oVirt, plain libvirt); VMware ESXi only through the NVMe/TCP gateway (docs.ceph.com, rbd/nvmeof-overview) or the iSCSI and NFS gateways; KubeVirt platforms through Rook and ceph-csi; no Hyper-V client found
    • Tier: Open-source engine
    • Archetype: Dependable builder
    • Evidence grade: B. The release table, the Squid and Tentacle release notes, the lifecycle policy, the foundation page and the licence file were read from docs.ceph.com, ceph.io and GitHub. What shipped in Hammer to Quincy (2015 to 2022) is dated from the release table but described from memory of the release notes, so the steps before 2023 are grade B and should be re-read before the next edition.

    Scope of the scores: Scored as the upstream project alone: what docs.ceph.com documents and what the release notes say shipped. Support contracts, certified hypervisor plug-ins, FIPS builds and appliances belong to the products built on it and are scored under IBM Storage Ceph, OpenShift Data Foundation, croit and Proxmox VE Ceph.

    Scorecard

    DimensionScoreDirection
    Capability today80.5 / 100higher is better
    Momentum since end of 2022+7.4 points (2015 to 2022: +19.7)higher is better
    Credibility19.0 / 25higher is better
    Commercial risknot scored (engine: nothing to buy)
    Technical lock-in7.0 / 25lower is better
    Integrator fit12.5 / 20higher is better
    AI leverage3.0 / 20higher is better
    Ledger Index67.4 / 100balanced weights; see the methodology

    Capability by domain, 2015 to today

    Scores 0 to 5 against fixed anchors; total is weighted to 100.

    Domain (weight)201520192022Today
    Data services core (15)2.53.53.54
    Resilience and HA (15)3.53.544.5
    Performance and protocols (12)2334
    Hypervisor integration (12)3333.5
    Scale and topology (10)44.54.54.5
    Data protection and DR (10)233.53.5
    Security (8)22.533.5
    Operations and automation (8)23.544
    Hardware and deployment freedom (5)5555
    Kubernetes and containers (5)0344
    Total / 10053.468.073.180.5

    Year by year: 2015: 53.4 · 2016: 55.4 · 2017: 58.4 · 2018: 59.2 · 2019: 68.0 · 2020: 68.8 · 2021: 71.3 · 2022: 73.1 · 2023: 75.5 · 2024: 77.0 · 2025: 80.5 · 2026: 80.5

    What moved the score

    YearDomainChangePointsTrigger
    2016Data protection and DR2 → 3+2.0Jewel: RBD mirroring between clusters (journal based)
    2017Data services core2.5 → 3.5+3.0Luminous: BlueStore default with inline compression; erasure-coded pools accept overwrites so RBD and CephFS can use them
    2018Operations and automation2 → 2.5+0.8Mimic: dashboard v2
    2019Kubernetes and containers0 → 3+3.0Rook 1.0 and ceph-csi: dynamic RBD and CephFS volumes with snapshots, expansion and ReadWriteMany file
    2019Operations and automation2.5 → 3.5+1.6Nautilus: dashboard with REST API, Prometheus module, PG autoscaler
    2019Performance and protocols2 → 3+2.4Nautilus: iSCSI gateway and NFS-Ganesha exports managed from the dashboard; all-flash BlueStore deployments documented
    2019Scale and topology4 → 4.5+1.0Nautilus: PG autoscaler and device classes make growth of mixed pools routine
    2019Security2 → 2.5+0.8Nautilus: msgr2 with on-wire encryption; dashboard roles
    2020Operations and automation3.5 → 4+0.8Octopus: cephadm orchestrator with rolling upgrades; Terraform and Ansible remain community-maintained, the data availability score is still a preview
    2021Data protection and DR3 → 3.5+1.0Pacific: stretch mode gives synchronous two-site protection; snapshot-based RBD mirroring (Octopus) and CephFS mirroring; no orchestrated failover, no immutable block snapshots, no backup vendor integration of its own
    2021Resilience and HA3.5 → 4+1.5Pacific: stretch mode for two-site clusters with a tiebreaker monitor
    2022Kubernetes and containers3 → 4+1.0Rook and ceph-csi: ReadWriteMany block for KubeVirt live migration, topology awareness, operator-based lifecycle; Kubernetes-native management is left to products such as OpenShift Data Foundation
    2022Security2.5 → 3+0.8Quincy: RBD per-image LUKS encryption; server-side encryption with an external KMS for object storage
    2023Hypervisor integration3 → 3.5+1.2Reef: NVMe/TCP gateway gives VMware ESXi a documented path; no vSphere plug-in, no VAAI statement and no Hyper-V client upstream
    2023Performance and protocols3 → 3.5+1.2Reef: NVMe/TCP gateway documented for ESXi and Linux initiators; upstream GA label not found, so counted as half a step
    2024Resilience and HA4 → 4.5+1.5Squid 19.2.0: MSR CRUSH rule type places erasure-coded chunks across failure domains more flexibly; 20.2.1 (2026-04-06) adds NVMe-oF gateway fast failover
    2025Data services core3.5 → 4+1.5Tentacle 20.2.0: FastEC partial reads and writes on erasure-coded pools (per-pool flag), improved BlueStore compression; still no deduplication and no per-volume minimum QoS
    2025Performance and protocols3.5 → 4+1.2Tentacle 20.2.0: SMB manager module with Samba on CephFS; NVMe/TCP gateway groups and multiple namespaces in the dashboard; no Fibre Channel, RDMA not mainstream, no independent benchmarks published by the project
    2025Security3 → 3.5+0.8Tentacle 20.2.0: oauth2-proxy single sign-on and certmgr for service certificates; 20.2.4 and Squid 19.2.6 (2026-08-19) fix the CephX authentication bypass CVE-2025-30156 and three more CVEs, and every key must be renewed. No FIPS validation, no anomaly detection, no immutable block snapshots

    Direction, 2023 to 2026

    Each release item carries one theme and one driver (V vision, C customer need, M market window, U upstream, P portfolio).

    DateItemThemeDriver
    2023-08Reef 18.2.0: NVMe/TCP gateway documented for block access from ESXi and LinuxPerformanceV
    2023-08Reef 18.2.0: RADOS and RBD performance work; release landed in August against a spring targetPackagingC
    2024-09Squid 19.2.0: MSR CRUSH rule type for flexible erasure-coded layoutsResilienceC
    2024-09Squid 19.2.0: Crimson and SeaStore first technology previewPerformanceV
    2024-09Squid 19.2.0: RGW user accounts with AWS-compatible IAMSecurityC
    2025-04Reef 18.2.5 to 18.2.7: hotfix for a critical BlueStore regression within a monthOperationsC
    2025-07Squid 19.2.3: RGW STS authentication bypass CVE-2023-43040 and JWT algorithm restriction CVE-2024-48916 fixedSecurityC
    2025-11Tentacle 20.2.0: FastEC partial reads and writes on erasure-coded poolsData servicesC
    2025-11Tentacle 20.2.0: SMB manager module with Samba on CephFS; NVMe/TCP gateway groups in the dashboardPerformanceC
    2025-11Tentacle 20.2.0: cephadm management gateway, oauth2-proxy single sign-on and certmgrSecurityC
    2025-11Tentacle 20.2.0: SeaStore with Crimson and the data availability score stay technology previewsOperationsV
    2026-03Reef 18.2.8: final release, Reef end of life on the published datePackagingU
    2026-04Tentacle 20.2.1: NVMe-oF gateway fast failover overhaul, RBD exclusive-transient lock modeResilienceC
    2026-06Tentacle 20.2.2: Rocky Linux 10 packagesPackagingC
    2026-06Umbrella (21.x) shown as in development at Ceph Days London, no datePackagingV
    2026-08Tentacle 20.2.3: configurable delayed NVMe-oF failback, NVMe-oF manager module, RGW mTLS for Kafka notificationsOperationsC
    2026-08Tentacle 20.2.4 and Squid 19.2.6: CephX authentication bypass CVE-2025-30156 and three more CVEs fixed; all keys must be renewedSecurityC
    2026-10Squid 19.2.7: backports only; Squid end of life estimated 2026-10-31PackagingU
    • Centre of gravity: Packaging (5), Security (4), Performance (3)
    • Blind spots (no items): Hypervisor integration, Scale, Data protection, Hardware, Kubernetes, Licensing, AI
    • Driver mix: Vision 22%, Customer need 67%, Market window 0%, Upstream 11%, Portfolio 0%
    • Idea-to-need index: 0.25 (0 = demand-pulled, 1 = thesis-pushed)

    Credibility: 19.0 / 25

    CriterionScoreBasis
    Cadence3.5Published annual cycle and four to six week point releases; the named release has landed between August and November since Quincy, and Squid went 11 months between 19.2.3 and 19.2.4
    Roadmap transparency4Open tracker, public release notes and developer summits; no dated roadmap, Umbrella has no published date
    Say-do3.5Features announced in release notes ship with the release; Crimson and SeaStore have stayed previews since 2024-09
    Velocity4.5Several substantive capabilities per named release: MSR rules and IAM accounts in Squid, FastEC, SMB and NVMe-oF failover in Tentacle
    Lifecycle stability3.5About two years of backports per series, upgrades tested from the last two releases, no downgrade; the Reef 18.2.5 regression needed a hotfix and the Squid point-release gap broke the stated rhythm

    Commercial risk: not scored

    This is an engine: nobody sells it, so there are no commercial terms to score.

    Technical lock-in: 7.0 / 25 (lower is better)

    CriterionScoreBasis
    Formats0.5Standard RBD images (export to raw or qcow2), POSIX file system, S3-compatible object
    Export path1.5rbd export, qemu-img and the NFS, SMB and NVMe/TCP gateways let data be copied out; no one-step export
    Stack coupling2.5Needs Ceph clients in the kernel or QEMU, or a gateway; VMware and Hyper-V reach it only through gateways
    Hardware / cloud coupling0Any x86 or Arm server, fully offline
    Skills and tooling coupling2.5Specialised operations skills; tooling (cephadm, Rook, dashboard) is open

    Integrator fit: 12.5 / 20

    CriterionScoreBasis
    Partner programme openness4Anyone may build, package and sell support on it; Ceph Foundation membership is open in three tiers; no vendor programme with margins or certification
    Multi-tenancy and self-service3Pools, RBD namespaces, CephFS subvolumes and RGW accounts with IAM (Squid); no portal or metering
    API and infrastructure-as-code3.5REST manager API and dashboard; Rook for Kubernetes; Terraform and Ansible content is community-maintained
    Skills and certification2No vendor-neutral certification found; training comes from the packaging vendors

    AI leverage: 3.0 / 20

    How the product uses AI, scored per year like capability. Only GA counts.

    Criterion20192022TodayLatest step
    AI workload infrastructure1112015: S3 (RGW) and CephFS serve as training data stores; no GPU data path, RDMA is not a documented mainstream option
    AI platform stack0002015: None
    AI-assisted operations0002015: None; the data availability score (Tentacle) is a rule-based preview
    Agent openness2222015: REST manager API with tokens and a CLI agents can drive; no official MCP server found

    Year by year: 2015: 3 · 2016: 3 · 2017: 3 · 2018: 3 · 2019: 3 · 2020: 3 · 2021: 3 · 2022: 3 · 2023: 3 · 2024: 3 · 2025: 3 · 2026: 3

    AI say-do, 2023 to 2026: 0 AI items announced; 0 GA, 0 still preview, 0 pending, 0 dropped. The project announced no AI feature between 2023 and 2026; a community blog post on using AI to brief remote hands (2026-10) is not a product feature.

    Ceph uses no AI in its own operation and offers none to operators; its role in AI is as the object and file store under other platforms. Agents can drive it through the REST API, and nothing more specific exists upstream.

    Reading

    Ceph scores as the deepest engine in this group because almost every capability a packaging vendor sells already exists upstream: erasure coding with FastEC, stretch mode, NVMe/TCP and SMB gateways, cephadm and Rook. What it lacks is exactly what the products add: a support contract, a vSphere plug-in, FIPS builds, a two-node option and anyone to call when the CephX keys need renewing. The security line sits at 3.5 because the 2026-08-19 fixes repaired a bypass in the authentication protocol itself and the project has no FIPS validation or anomaly detection. The one thing to check before relying on it is which stable series your packaging vendor ships and how far behind upstream its point releases run, because Squid reaches end of life around 2026-10-31.

    Open questions and items to re-score

    • Whether the NVMe-oF gateway carries a technology preview or GA label upstream was not found on docs.ceph.com; it is counted as half a step from Reef.
    • The 11 month gap between Squid 19.2.3 (2025-07-28) and 19.2.4 (2026-06-01) in the release table should be confirmed before quoting.
    • Umbrella (21.x) has no published date; the keynote PDF could not be read.
    • The current Premier member roster of the Ceph Foundation was not read.
    • Capability steps for Hammer to Quincy (2015 to 2022) are described from memory of the release notes and should be re-read.
    • Whether a vendor-neutral Ceph certification exists was not confirmed.

    Related reading

    More from Virtualization Storage Ledger 2026