Skip to content

OpenNebula

Open-source cloud management platform for building sovereign, hybrid, and edge clouds — a lightweight alternative to OpenStack with native AI factory support.

Overview

OpenNebula is a unified cloud management platform designed for simplicity, operational efficiency, and full-stack management of VMs, containers, and bare metal. It provides a centralized architecture that is significantly easier to deploy and operate than OpenStack, while supporting enterprise-grade features including NVIDIA GPU orchestration, live migration, and hybrid/edge cloud management. It positions itself as the leading VMware alternative for organizations seeking sovereignty and cost control.

Repository & Community

Attribute Detail
Repository github.com/OpenNebula/one
Stars ~1.5k ⭐
Latest Version v7.2 (April 7, 2026)
Language C++, Ruby, JavaScript
License Apache 2.0
Company OpenNebula Systems

Evaluation

  • Why it's better: 10x simpler than OpenStack to deploy and operate. Centralized architecture (oned) vs distributed services. Native NVIDIA GPU orchestration for AI factories. Native LXC containers alongside VMs. Production-ready in hours, not weeks. Full VMware migration tooling.

  • When it fits (Applicability):

  • VMware replacement (post-Broadcom)
  • SMB/mid-size private cloud
  • Edge computing with distributed locations
  • AI/GPU factories (NVIDIA Grace Blackwell, NVSwitch)
  • Hybrid cloud (on-prem + public cloud bursting)
  • Research institutions and universities

  • Managed service providers

  • Pros and Cons:

Pros Cons
Dramatically simpler than OpenStack Smaller community than OpenStack/K8s
Centralized, easy to troubleshoot Fewer third-party integrations
Native NVIDIA GPU orchestration Less battle-tested at hyperscale
VMware migration tools Smaller talent pool
Apache 2.0, no vendor lock-in UI less polished than commercial alternatives
KVM + LXC + K8s in one platform Enterprise features require subscription
Edge-native architecture Less documentation than OpenStack
gRPC API for high-concurrency (v7.2)

Architecture

flowchart TB
    subgraph Frontend["OpenNebula Front-End"]
        ONED["oned\n(Core Daemon)"]
        Sched["Scheduler\n(placement engine)"]
        OneDRS["OneDRS\n(AI-powered DRS)"]
        Sunstone["Sunstone\n(Web UI)"]
        OneGate["OneGate\n(VM contextualization)"]
        OneFlow["OneFlow\n(service orchestration)"]
        FireEdge["FireEdge\n(Guacamole VNC/SSH)"]
        gRPC_API["gRPC API\n(v7.2+)"]
        XML_RPC["XML-RPC API\n(legacy)"]
    end

    subgraph Hosts["Compute Hosts"]
        KVM["KVM\n(VM hypervisor)"]
        LXC["LXC\n(system containers)"]
        FC["Firecracker\n(microVMs)"]
    end

    subgraph Storage["Storage"]
        FS["Filesystem\n(NFS, GlusterFS)"]
        LVM_S["LVM"]
        Ceph_S["Ceph\n(RBD)"]
        SAN["SAN / NetApp /\nPure Storage"]
    end

    subgraph Network["Networking"]
        Bridge["Linux Bridge"]
        VXLAN["VXLAN"]
        OVS_ON["Open vSwitch"]
        SRIOV["SR-IOV"]
    end

    subgraph DB["Database"]
        MySQL["MySQL / MariaDB"]
        SQLite["SQLite\n(small deployments)"]
    end

    ONED --> Hosts
    ONED --> Storage
    ONED --> Network
    ONED --> DB
    Sched --> ONED
    Sunstone --> ONED
    gRPC_API --> ONED

    style Frontend fill:#00758f,color:#fff
    style Hosts fill:#1565c0,color:#fff

Key Features

Feature Detail
VM Management Full lifecycle: create, migrate, snapshot, resize
LXC Containers Production-grade system containers with VM-like management
NVIDIA GPU Orchestration NVSwitch, NVLink, Fabric Manager, BlueField DPU
Live Migration Cross-host, cross-datastore, EVC for heterogeneous CPUs
OneDRS AI-powered Distributed Resource Scheduler
Hybrid Cloud Burst to AWS, Azure, GCP, Equinix
Edge Computing Distributed cluster management across locations
OneFlow Multi-VM service orchestration
OneForm Automated cluster deployment
Marketplace Pre-built VM/container appliances
RBAC & Multi-tenancy Groups, ACLs, resource quotas
Sunstone UI Real-time VM logs, visual management

v7.2 Highlights (April 2026)

Feature Detail
gRPC API Low-latency, high-throughput communication
NVIDIA Grace Blackwell GB200/GB300 validated support
Spectrum-X networking NVIDIA networking platform integration
BlueField DPU Hardware-level network offloading
Storage mobility Live migration between LVM and file datastores
IP sharing VM group IP address sharing
EVC Enhanced Vmotion Compatibility across CPU gens
Everpure FlashArray Native storage driver
NetApp incremental backup Incremental backup support
RHEL 10 / AlmaLinux 10 / Debian 13 OS support

Pricing

Tier Cost Notes
Community (CE) Free (Apache 2.0) Full platform, community support
Enterprise (EE) Subscription-based Long-term support, patches, SLA
Support From €6,000/yr Standard, Premium, Priority tiers
Professional Services Custom Implementation, migration, training

Compatibility

Dimension Support
Hypervisors KVM (primary), LXC, Firecracker, VMware (migration)
Storage Ceph, NFS, LVM, iSCSI, GlusterFS, NetApp, Pure Storage
Networking Linux Bridge, VXLAN, Open vSwitch, SR-IOV, 802.1Q
Cloud providers AWS, Azure, GCP, Equinix Metal
GPU NVIDIA (NVSwitch, NVLink, BlueField, CUDA passthrough)
OS Ubuntu 22/24, RHEL 8/9/10, AlmaLinux 8/9/10, Debian 11/12/13, SUSE 15
CPU architecture amd64, arm64

Sources

Source URL Retrieved Via
Official Website https://opennebula.io Direct
Documentation https://docs.opennebula.io Direct
GitHub Repository https://github.com/OpenNebula/one Direct
v7.2 Release Notes https://github.com/OpenNebula/one/releases/tag/release-7.2.0 Web Search
Architecture Docs https://docs.opennebula.io/7.2/overview/architecture.html Direct
Blog https://opennebula.io/blog/ Direct
Case Studies https://opennebula.io/case-studies/ Web Search
Community https://forum.opennebula.io Direct
Marketplace https://marketplace.opennebula.io Direct

Questions

Open Questions

Answered Questions

  • How does the gRPC API (v7.2) compare to the legacy XML-RPC API in terms of latency and throughput at 1,000+ concurrent operations? — The gRPC API (introduced v7.2) uses Protocol Buffers over HTTP/2 with multiplexed streams, compared to XML-RPC's single-request-per-connection HTTP/1.1 model. Expected improvements: lower per-request latency (binary vs text serialization), higher throughput (HTTP/2 multiplexing avoids connection overhead), and better streaming support. No published benchmarks at 1,000+ concurrent operations. The oned daemon remains single-threaded, so the API protocol improvement helps serialization overhead but the daemon itself remains the throughput bottleneck at extreme concurrency. Federation is recommended for deployments exceeding 500 hosts. — resolved via OpenNebula v7.2 release notes

  • How does OneDRS AI placement compare to VMware DRS in production environments? — No published direct benchmark comparison exists. VMware DRS is mature with rule-based + predictive AI placement (vSphere 8+), hierarchical resource pools, and granular affinity/anti-affinity. OpenNebula's scheduler (pluggable, default "freedom") supports custom placement algorithms, affinity rules, and live migration but lacks the deep ML-driven workload prediction of VMware DRS. OpenNebula targets VMware replacement use cases with explicit migration tooling. For basic load balancing and policy-based placement, OpenNebula is sufficient; for complex multi-tier application SLO management, VMware DRS remains more sophisticated. — resolved via OpenNebula and VMware documentation

  • What is the maximum tested cluster size for OpenNebula (number of hosts and VMs)? — OpenNebula has been tested up to 1,000+ hosts and 30,000+ VMs in a single cluster. The oned daemon is single-threaded, so the XML-RPC API can become a bottleneck at very high scale. Federation (multiple OpenNebula zones) is recommended for deployments exceeding 500 hosts. — resolved via OpenNebula documentation

  • What is the production maturity of LXC support in v7.2? — LXC support in OpenNebula is functional but considered less mature than KVM. It supports basic VM lifecycle, templates, and networking, but lacks some advanced features available with KVM (live migration, snapshots with certain storage backends). Best suited for lightweight workloads and development environments. For production, KVM remains the recommended hypervisor. — resolved via OpenNebula documentation

  • Is OpenNebula truly simpler than OpenStack? → Yes — single daemon (oned) vs 30+ distributed services. Installations take hours vs weeks.

  • Does OpenNebula support NVIDIA GPUs? → Yes — native NVSwitch, NVLink, Fabric Manager, BlueField DPU support in v7.2. See infrastructure/opennebula/index#v7.2 Highlights (April 2026).
  • Can OpenNebula replace VMware? → Yes — explicit VMware migration tooling and positioning as a VMware alternative. See infrastructure/opennebula/index.
  • What hypervisors are supported? → KVM (primary), LXC, Firecracker. VMware for migration only.