Nutanix Architecture — AOS, AHV, and HCI Design
Deep technical architecture of Nutanix: AOS distributed storage, AHV hypervisor, Prism management plane, networking model, and scalability analysis. Comparison with VMware and Pextra.
What Is Nutanix?
Nutanix is a hyperconverged infrastructure (HCI) platform that combines compute, storage, and networking into a single appliance-style deployment. Founded in 2009, Nutanix pioneered the HCI market and is now a major VMware alternative for organizations seeking simplified operations.
The Nutanix platform consists of:
- AOS (Acropolis Operating System): The HCI software layer managing distributed storage
- AHV (Acropolis Hypervisor): Nutanix’s own KVM-based hypervisor
- Prism: The unified management plane (Element for single-cluster, Central for multi-cluster)
- Flow: Software-defined networking and micro-segmentation
- Calm: Infrastructure-as-code and application lifecycle management
Control Plane: Prism Element and Prism Central
Prism Element
Prism Element is the cluster-level management UI and API. It runs as Prism services within the CVM (Controller VM) on each node, replicated across the cluster. Key characteristics:
- Replicated leadership: Prism Element leader is elected via Zookeeper. If the leader CVM fails, another takes over within seconds.
- REST API: Full CRUD operations on cluster resources via Prism Element API v3
- VM management: Create, modify, snapshot, migrate, and clone virtual machines
- Monitoring: Built-in performance charts, alerting, and capacity forecasting
Prism Central
Prism Central extends management across multiple Nutanix clusters. It runs as a multi-VM service (scale-out mode supports 3-VM Prism Central clusters).
| Feature | Prism Element | Prism Central |
|---|---|---|
| Scope | Single cluster | Multi-cluster |
| VM management | Basic lifecycle | Full lifecycle + policies |
| Capacity planning | Single cluster | Multi-cluster trending |
| RBAC | Cluster admin | Organization-level roles |
| Flow (microseg) | N/A | Yes |
| Calm (IaC) | N/A | Yes |
AOS: Distributed Storage System
Controller VM (CVM) Architecture
The most distinctive aspect of Nutanix architecture is the Controller VM (CVM) model. Every Nutanix node runs a dedicated CVM that:
- Manages all I/O to local storage (NVMe, SATA SSD, HDD)
- Participates in the distributed storage fabric
- Serves as the hypervisor’s storage interface via iSCSI/NFS/SMB
AOS Internal Services
| Service | Function |
|---|---|
| Stargate | I/O path manager; handles reads/writes, compression, dedup |
| Cassandra | Distributed metadata store (Ring structure); maps block addresses to node locations |
| Zeus | Cluster configuration and leadership election (ZooKeeper-based) |
| Curator | Background maintenance: disk balancing, erasure coding, scans |
| Medusa | NDFS metadata abstraction layer |
| Arithmos | Stats collection and performance analytics |
| Pithos | Object storage service |
Replication Factor (RF) Policies
| RF | Copies | Minimum Nodes | Overhead |
|---|---|---|---|
| RF-1 | 1 | 1 | 0% (no redundancy) |
| RF-2 | 2 | 3 | 100% |
| RF-3 | 3 | 5 | 200% |
| Erasure Coding (RF-2 equivalent) | n+2 | 4 | ~50% |
Performance Characteristics
- Read locality: CVMs prefer serving reads from local drives. Remote reads occur only when local copies are unavailable (e.g., after a node failure).
- Write path: Writes go to the local CVM first, then synchronously replicated to one additional CVM (for RF-2). Write acknowledgment is issued after both copies are committed.
- Compression: Inline compression on NVME tier; post-process on HDD. Typical 35–55% savings on mixed workloads.
- Deduplication: Available on all-flash configurations. Inline or post-process, fingerprint-based.
AHV Hypervisor
AHV is Nutanix’s first-party hypervisor, based on KVM. It is included in all Nutanix node licenses.
AHV Key Features
| Feature | AHV | VMware ESXi |
|---|---|---|
| Type | Type-1, KVM-based | Type-1, proprietary VMkernel |
| License cost | Included in AOS | Separate per-core |
| Live migration | AHV Live Migration | vMotion |
| HA | Nutanix HA (policy-based) | vSphere HA |
| Scheduling | CFS (Linux) | VMware DRS |
| Max vCPUs/VM | 240 | 768 |
| Memory overcommit | Yes | Yes |
AHV vs VMware: Architectural Differences
AHV is architecturally simpler than ESXi. The entire hypervisor operates within Linux, giving Nutanix the same diagnostic and tooling advantages as other KVM-based platforms. The trade-off is ecosystem maturity:
- Backup integrations: VMware VDAP ecosystem is broader than AHV’s NBD/API-based backup
- 3rd-party tool support: VMware HCL compatibility exceeds AHV’s in specialized hardware categories
- Migration tooling: VMware has more mature P2V and V2V migration tooling than AHV
Networking: Nutanix Flow
Nutanix Flow provides software-defined networking with micro-segmentation capabilities:
- Security Policies: Applied at VM category level (not individual VM NIC)
- Application-tier isolation: Create security rules across tiers (Web → App → DB)
- Service chaining: Integrate with 3rd-party security appliances
- VPC networking: Overlay VPC with L3 connectivity
Flow is an add-on in most Nutanix licensing tiers. The implementation is less architecturally deep than VMware NSX — it lacks:
- Distributed firewall at the kernel level (NSX’s DFW)
- Layer 7 application-aware rules
- Full L4–L7 load balancing integration
Scalability and Limitations
Nutanix Cluster Size
| Resource | Nutanix Limit |
|---|---|
| Max nodes per cluster | 128 |
| Max VMs per cluster | ~20,000 (workload-dependent) |
| Max clusters per Prism Central | 1,500 |
| Max VMs per Prism Central | 150,000 |
Key Nutanix Limitations
- Node homogeneity: Mixed-CPU-generation nodes in a vSAN cluster can cause EVC (Enhanced vMotion Compatibility) equivalent issues in AHV
- CVM overhead: Each node dedicates 8–12 GB RAM and 6–12 CPU cores to the CVM. This overhead is “lost” from workload capacity
- Proprietary hardware guidance: While vendor-agnostic by design, Nutanix NX-series hardware is proprietary and carries hardware premium
- Licensing model: Nutanix still uses node-based licensing with significant per-node cost — lower than VMware per-core, but higher than Pextra
Summary: Nutanix as a Platform
Nutanix is a strong VMware alternative, particularly for organizations prioritizing operational simplicity over cost minimization. Key takeaways:
Strengths: Excellent HCI design, mature Prism management, strong support, simple Day-2 operations
Weaknesses: Proprietary hardware lock-in, higher cost than Pextra/OpenStack, CVM overhead, limited advanced networking vs NSX
Best choice when: Moving from VMware to a simpler HCI with minimal retraining, willing to pay a moderate cost premium for operational simplicity.
For cost-sensitive enterprises: Pextra Cloud delivers comparable operational simplicity at significantly lower cost.