Skip to content

OpenStack

The industry-standard open-source cloud operating system for building and managing large-scale private and public cloud infrastructure (IaaS).

Overview

OpenStack is a modular, service-oriented cloud platform that treats the data center as a single programmable resource pool. It provides APIs and tools for managing compute, storage, and networking resources at massive scale. Governed by the OpenInfra Foundation (formerly OpenStack Foundation), it is deployed by thousands of organizations worldwide, from telecom providers to research institutions to enterprises replacing VMware.

Repository & Community

Attribute Detail
Repository opendev.org/openstack
Latest Version 2026.1 "Gazpacho" (April 1, 2026) — SLURP release
Language Python
License Apache 2.0
Governance OpenInfra Foundation
Contributors 10,000+ (all time)

Evaluation

  • Why it's better: The most mature, feature-complete open-source IaaS platform. Full-stack infrastructure management (compute, network, storage, identity, images, orchestration). Massive community, proven at hyperscale (100k+ cores). Key VMware replacement in the post-Broadcom era.

  • When it fits (Applicability):

  • Large-scale private cloud (1,000+ nodes)
  • Telecom NFV infrastructure
  • VMware replacement / migration
  • Multi-tenant cloud service provider
  • Bare metal provisioning (via Ironic)

  • Scientific computing / HPC clusters

  • Pros and Cons:

Pros Cons
Most mature open-source IaaS Very high operational complexity
Apache 2.0, no vendor lock-in Requires dedicated platform team
Massive community & ecosystem Many moving parts (30+ projects)
Proven at hyperscale Steep learning curve
VMware migration path Eventlet → async migration ongoing
SLURP releases for stability Upgrade complexity between releases
Bare metal via Ironic Heavy resource requirements

Architecture

flowchart TB
    subgraph ControlPlane["Control Plane"]
        KS["Keystone\n(Identity & Auth)"]
        Glance["Glance\n(Image Service)"]
        Nova_API["Nova API\n(Compute)"]
        Neutron_API["Neutron\n(Networking)"]
        Cinder_API["Cinder\n(Block Storage)"]
        Heat["Heat\n(Orchestration)"]
        Horizon["Horizon\n(Web Dashboard)"]
    end

    subgraph Infra["Infrastructure Services"]
        MQ["RabbitMQ\n(Message Queue)"]
        DB["MariaDB / Galera\n(Database)"]
        Memcache["Memcached\n(Token cache)"]
    end

    subgraph Compute["Compute Nodes"]
        Nova_C["nova-compute\n(KVM / QEMU)"]
        OVN_A["OVN Agent\n(networking)"]
    end

    subgraph Storage["Storage"]
        Ceph["Ceph\n(block, object, file)"]
        Swift["Swift\n(Object Storage)"]
    end

    subgraph Network["Network Nodes"]
        OVN_C["OVN Controller\n(SDN)"]
        LB["Octavia\n(Load Balancer)"]
    end

    KS <--> MQ
    Nova_API <--> MQ <--> Nova_C
    Neutron_API <--> MQ <--> OVN_A
    Cinder_API <--> MQ
    All <--> DB
    Nova_C --> Ceph
    Cinder_API --> Ceph

    style ControlPlane fill:#ef3e42,color:#fff
    style Compute fill:#1565c0,color:#fff

Core Services

Service Project Role
Identity Keystone Authentication, authorization, service catalog
Compute Nova VM lifecycle, scheduling, live migration
Networking Neutron SDN, OVN, BGP, security groups, floating IPs
Block Storage Cinder Persistent volumes, snapshots, encryption
Object Storage Swift S3-compatible object store
Image Glance VM image registry and distribution
Dashboard Horizon Web GUI for operators and tenants
Orchestration Heat Stack-based resource orchestration (CFN-compatible)
Load Balancer Octavia LBaaS (amphora, OVN)
Bare Metal Ironic Physical server provisioning
DNS Designate DNSaaS
Container Magnum K8s cluster lifecycle on OpenStack

2026.1 "Gazpacho" Highlights

Feature Detail
Parallel live migration Multiple memory transfer connections simultaneously
Live migration with vTPM No cold restart for security-sensitive VMs
OVN BGP support Native BGP for external connectivity
Default IOThread Per-QEMU instance for disk I/O offloading
Async volume attach Decouple storage ops from API response
Ironic improvements Auto NFS/CIFS detection, trait-based port scheduling
SLURP release Direct upgrade from 2025.1 "Epoxy"

Pricing

Offering Cost
Self-hosted Free (Apache 2.0)
Canonical/MAAS Support subscriptions
Red Hat OpenStack Platform Enterprise subscription
SUSE Cloud Enterprise subscription
Managed (Rackspace, Vexxhost) Per-resource pricing

Compatibility

Dimension Support
Hypervisors KVM (primary), QEMU, Xen, VMware vCenter, Hyper-V, Ironic (bare metal)
Storage Ceph (block/object/file), LVM, NFS, iSCSI, FC, NetApp, Pure Storage
Networking OVN (default), OVS, Linux Bridge, SR-IOV, Mellanox
OS Ubuntu, RHEL, CentOS Stream, Rocky, Debian, SUSE
CPU architecture amd64, arm64, ppc64le, s390x

Sources

Source URL Retrieved Via
Official Website https://openstack.org Direct
Documentation https://docs.openstack.org Direct
Releases https://releases.openstack.org Direct
Gazpacho 2026.1 https://releases.openstack.org/gazpacho/ Web Search
OpenDev https://opendev.org/openstack Direct
Nova Docs https://docs.openstack.org/nova/latest/ Direct
Neutron Docs https://docs.openstack.org/neutron/latest/ Direct
Cinder Docs https://docs.openstack.org/cinder/latest/ Direct
Keystone Docs https://docs.openstack.org/keystone/latest/ Direct
OpenInfra Foundation https://openinfra.dev Direct
Sardina Systems Review https://sardinasystems.com Web Search

Questions

Open Questions

Answered Questions

  • How does parallel live migration in Gazpacho perform under 10+ concurrent VM migrations? — Nova supports parallel live migrations via max_concurrent_live_migrations per compute host. Under 10+ concurrent migrations, performance is dominated by network bandwidth and storage I/O contention rather than Nova itself. Key tuning: live_migration_bandwidth (MiB/s), live_migration_completion_timeout, and ensuring sufficient network capacity (dedicated migration VLAN recommended). No published benchmarks specific to 2026.1 Gazpacho. Best practice: limit concurrent migrations per host to 2-4 for production stability, use shared storage to minimize data transfer. — resolved via Nova live migration documentation

  • How does Magnum (K8s on OpenStack) compare to Cluster API for K8s lifecycle management? — Magnum is OpenStack-native, managing K8s via Heat templates with deep OpenStack integration (Neutron, Cinder, Keystone). Cluster API (CAPI) is Kubernetes-native (CRD-based, declarative, GitOps-friendly) with a CAPO provider for OpenStack. Key trade-offs: Magnum is OpenStack-only and can lag on K8s version support; CAPI is multi-cloud, generally faster to support new K8s versions, and integrates with Flux/ArgoCD. Industry trend is migration from Magnum to CAPI+CAPO for organizations wanting consistent multi-cloud K8s management. Magnum remains viable for OpenStack-only deployments wanting a single management plane. — resolved via OpenStack Magnum and Cluster API documentation

  • What is the status of the Eventlet to asyncio migration? Which services have completed the transition? — The Eventlet-to-asyncio migration is an ongoing multi-year effort (Project "oslo.async"). As of 2026.1, several services have made significant progress: Keystone, Glance, and designate have partial asyncio support. Nova and Neutron still rely heavily on Eventlet. The migration is complex because Eventlet's monkey-patching affects all thread and IO operations. Full migration is expected over the next 2-3 release cycles. — resolved via OpenStack release notes and community discussions

  • What is the recommended Ceph version for OpenStack 2026.1 Gazpacho? — Ceph Squid (v19.x) is the recommended version for OpenStack 2026.1. Ceph Reef (v18.x) is also fully supported. The RBD integration with Cinder and Glance is tested against both. For new deployments, use Squid. For upgrades from 2025.1 (Caracal), ensure Ceph is at minimum Reef before proceeding with the OpenStack upgrade. — resolved via OpenStack 2026.1 release notes

  • What is the latest release? → 2026.1 "Gazpacho" (April 1, 2026). SLURP release. See infrastructure/openstack/index#2026.1 Gazpacho Highlights.

  • What upgrade paths are supported? → SLURP-to-SLURP (2025.1 → 2026.1) or sequential (2025.2 → 2026.1).
  • Does OpenStack support bare metal? → Yes, via Ironic. Provisions physical servers with PXE/Redfish.
  • What is the minimum deployment? → 3 nodes (1 controller, 2 compute) for a basic cloud. Single-node possible with DevStack.