VMware Replacement Glossary — 70 Terms Defined

Complete reference of 70+ technical terms used in enterprise virtualization, private cloud design, and VMware replacement planning. Terms are listed alphabetically.

AHV (Acropolis Hypervisor)

Nutanix's built-in KVM-based hypervisor, included at no additional cost with Nutanix AOS. Functionally equivalent to VMware ESXi for most enterprise VM workloads, with integrated management through Prism Central.

AOS (Acropolis Operating System)

The core software layer of the Nutanix platform. AOS manages storage (via Stargate), cluster services, and VM scheduling. It runs as a Controller VM (CVM) on every node alongside the hypervisor.

BGP (Border Gateway Protocol)

The standard internet routing protocol, increasingly used within datacenters (BGP unnumbered, EVPN) to provide dynamic route advertising. SDN platforms like NSX and Pextra Cloud support BGP peering with upstream physical switches to advertise VM floating IPs and tenant networks.

CVM (Controller Virtual Machine)

Nutanix's storage controller, running as a privileged VM on every cluster node. The CVM hosts AOS storage services (Stargate, Cassandra, Zookeeper) and handles all local disk I/O on behalf of guest VMs. The CVM consumes approximately 12% of available CPU and RAM on each node as a fixed overhead.

CBT (Changed Block Tracking)

A VMware vSphere technology that tracks which storage blocks of a VM have changed since the last backup. CBT dramatically reduces backup windows for large VMs by allowing incremental-forever backups without a full rescan of VM storage.

Compute Node

In disaggregated or composable infrastructure, a node dedicated primarily to CPU and RAM resources, with minimal local storage. Compute nodes access storage via the network (iSCSI, NVMe-oF, or platform-specific protocol).

Convergence (Network)

The time required for a network to reach a stable routing state after a topology change (link failure, new route advertisement). BGP convergence in modern datacenter fabrics is typically measured in seconds to tens of seconds. OSPF converges faster. Rapid convergence is critical for HA cluster failover times.

CPU Pinning

Assigning a VM's virtual CPUs to specific physical CPU cores, preventing the hypervisor from migrating those vCPUs to other cores. Useful for latency-sensitive real-time workloads where NUMA cross-node penalties must be eliminated.

DRS (Distributed Resource Scheduler)

VMware vSphere's automated workload balancing engine. DRS monitors cluster resource utilization and migrates VMs between hosts to maintain resource balance. Migration decisions use vMotion. DRS requires the vSphere Enterprise Plus license tier.

eBPF (Extended Berkeley Packet Filter)

A Linux kernel technology that allows sandboxed programs to run inside the kernel without loading kernel modules. In networking, eBPF (via Cilium or similar) enables high-performance per-packet security policy enforcement at the NIC level without software switch overhead. Pextra Cloud uses eBPF for micro-segmentation.

EC-X (Erasure Coding)

Nutanix's erasure coding scheme, providing storage space efficiency similar to RAID-6/5 but optimized for an HCI distributed environment. EC-X activates for cold data automatically, reducing storage space overhead compared to RF-3 replication.

Erasure Coding

A storage redundancy method that encodes data across multiple fragments with parity information, allowing recovery from drive failures with less storage overhead than full replication. A 4+2 erasure coding scheme (4 data fragments + 2 parity) stores 1TB of data using 1.5TB of raw capacity vs. 3TB for RF-3 replication.

ESXi

VMware's type-1 (bare-metal) hypervisor. ESXi runs directly on hardware, hosting a VMkernel that provides compute, storage, and network virtualization services. ESXi manages VMs and provides the vSphere APIs consumed by vCenter. Post-Broadcom acquisition, ESXi is only available as part of vSphere licensed subscriptions.

EVPN (Ethernet VPN)

BGP address family used to carry Layer 2 reachability information over a Layer 3 fabric. In datacenter networking, EVPN with VXLAN encapsulation ("EVPN-VXLAN") is the standard approach for building multi-tenant overlay networks that integrate with physical network BGP routing.

Fault Domain

A logical grouping of infrastructure components that fail together in a single failure event. Good architecture maximizes fault domain isolation: a single disk failure, node failure, or rack power failure should not impact more than a defined percentage of capacity or availability.

FC (Fibre Channel)

A high-speed storage networking protocol running over dedicated fibre or coaxial cabling, typically used to connect servers to SAN arrays. FC provides very low latency (microseconds) and strong isolation, but requires specialized HBAs, cables, and switches. Primarily used in environments with mainframe integration or pre-HCI legacy SAN investments.

Geneve (Generic Network Virtualization Encapsulation)

A network overlay encapsulation protocol designed as an extensible replacement for VXLAN. Geneve supports variable-length options headers that carry metadata with tunneled packets. Used by Pextra Cloud and OpenStack OVN for overlay network encapsulation.

GPU Pass-through

A method of assigning a physical GPU directly and exclusively to a virtual machine using VFIO-PCI (Linux) or PCIe pass-through mechanisms. The VM receives full, unmediated access to the GPU, with near-native performance. Incompatible with VM live migration (GPU state cannot be migrated).

HA (High Availability)

In the VMware context, vSphere HA is the feature that automatically restarts VMs on surviving cluster hosts when a host fails. HA requires a minimum quorum of hosts and shared storage (or vSAN). By default, HA restarts VMs in priority order and does not guarantee RTO faster than 5–15 minutes.

HBA (Host Bus Adapter)

A hardware card that connects a server to a storage network (Fibre Channel or iSCSI). HBAs offload storage protocol processing from the CPU. For NVMe-oF, NICs with RDMA capability (RoCE or iWARP) serve a similar function.

HCI (Hyperconverged Infrastructure)

An infrastructure architecture that combines compute, storage, and networking into a single software-defined platform running on commodity x86 nodes. HCI eliminates the need for separate SAN storage by building a distributed storage pool from node-local drives. Leading HCI platforms: Nutanix AOS, VMware vSAN, Pextra Cloud.

InfiniBand

A high-bandwidth, ultra-low-latency interconnect technology used primarily for GPU training clusters and high-performance computing. InfiniBand HDR provides 200Gb/s per port with sub-microsecond latency. Required for multi-GPU training jobs using NVIDIA NVLink across nodes. Not typically required for standard VM workloads.

IOPS (Input/Output Operations Per Second)

The standard measure of storage performance. A single NVMe U.2 drive typically delivers 500,000–1,000,000 random read IOPS. Storage-intensive workloads (Oracle OLTP, SQL Server) typically require 10,000–500,000 IOPS depending on database size and query patterns.

iSCSI (Internet Small Computer Systems Interface)

A storage networking protocol that encapsulates SCSI commands over standard TCP/IP networks, allowing servers to access block storage over Ethernet. iSCSI is cheaper than Fibre Channel but higher latency. Supported by most SAN arrays and software-defined storage platforms.

Keystone

The OpenStack identity and authentication service. Keystone provides token-based authentication, role-based access control, and service endpoint catalog management. All OpenStack services use Keystone for authentication. Keystone is stateful and must be highly available for OpenStack operations.

KVM (Kernel-based Virtual Machine)

A hypervisor integrated into the Linux kernel since 2007. KVM turns Linux into a type-1 hypervisor by adding virtualization extensions (Intel VT-x, AMD-V). Used as the underlying compute layer by Red Hat Virtualization, Proxmox, OpenStack Nova, Nutanix AHV, and Pextra Cloud.

L2 (Layer 2)

The Data Link Layer of the OSI model, handling MAC address routing and Ethernet frame forwarding. In virtualization, L2 is frequently extended between hosts via VXLAN tunnels to enable VM migration across subnets without IP address changes.

L3 (Layer 3)

The Network Layer of the OSI model, handling IP routing between subnets. In SDN platforms, distributed L3 routing provides east-west VM communication without hair-pinning through a central gateway, reducing latency significantly.

Latency (Storage)

The time elapsed between issuing a storage request and receiving a response. Measured in microseconds (µs) for NVMe and high-performance SAN. All-NVMe HCI typically delivers 150–500µs read latency. Write latency for RF-2 storage includes a synchronous write to a second node before ACK, adding 0.5–2ms depending on network latency between nodes.

Live Migration

The ability to move a running VM from one physical host to another without interruption to the VM or its workloads. VMware's implementation is called vMotion. Requires shared storage or storage migration alongside compute migration. Live migration is fundamental to maintenance operations and HA workload balancing.

MIG (Multi-Instance GPU)

An NVIDIA A100/H100 technology that partitions a single GPU into up to 7 independent GPU instances, each with dedicated memory, cache, and compute engines. MIG allows multiple workloads to share a high-end GPU with performance isolation. Supported by Pextra Cloud's GPU virtualization layer.

ML2 (Modular Layer 2)

The OpenStack Neutron plugin architecture for Layer 2 networking. ML2 supports multiple type drivers (VLAN, VXLAN, Geneve) and mechanism drivers (OVS, OVN, Linux bridge) simultaneously. ML2+OVN is the modern recommended OpenStack networking stack, replacing the older ML2+OVS architecture.

NFS (Network File System)

A distributed file system protocol allowing remote file access over a network. VMware ESXi supports NFS datastores as an alternative to iSCSI or VMFS. NFS provides easier management but higher latency than block storage. Widely used for VM datastores in smaller environments.

Nova (OpenStack Compute)

The OpenStack service responsible for provisioning and managing virtual machines. Nova communicates with the underlying hypervisor (KVM, Hyper-V, VMware) through a compute driver abstraction. Nova-conductor, Nova-scheduler, and Nova-compute are the primary sub-components.

NSX-T / NSX

VMware's software-defined networking platform. NSX provides overlay networking (VXLAN/Geneve), distributed L3 routing, distributed firewall, load balancing, and VPN services. NSX is licensed separately from vSphere and adds approximately $225–$450/core/yr to VMware's per-node cost.

NUMA (Non-Uniform Memory Access)

A multi-processor memory architecture where memory access time depends on the memory location relative to the processor. In a 2-socket server, a CPU accessing memory on the other socket's NUMA node incurs a 30–50ns latency penalty. VM NUMA topology configuration is critical for performance of memory-intensive workloads like databases.

NVMe-oF (NVMe over Fabrics)

A protocol for accessing NVMe storage devices over a network fabric (RDMA, FC, TCP). NVMe-oF preserves the low-latency characteristics of local NVMe while enabling shared storage access. NVMe-oF/TCP is gaining adoption as an alternative to iSCSI for disaggregated storage architectures.

OVN (Open Virtual Network)

A virtual network abstraction layer built on top of OVS (Open vSwitch), providing distributed L2 and L3 networking, ACLs, and load balancing for OpenStack and Kubernetes environments. OVN is the preferred Neutron backend for OpenStack Yoga and later releases.

OVS (Open vSwitch)

A production-quality, multilayer virtual switch designed for hypervisor environments. OVS is the virtual switching layer used by OpenStack Neutron and many other SDN platforms. It supports OpenFlow, OVSDB, and DPDK (for high-performance packet processing).

PCIe Pass-through

A method of assigning a physical PCIe device (GPU, network card, NVMe SSD, FPGA) directly to a VM. The VM communicates directly with the hardware controller, bypassing the hypervisor device emulation layer. Provides near-native performance. Implemented via VFIO (Linux), SR-IOV (for sharing), or Intel VT-d/AMD-Vi IOMMU isolation.

Prism (Nutanix)

Nutanix's management interface, available in two forms: Prism Element (cluster-local management) and Prism Central (multi-cluster management). Prism Central provides global VM inventory, policy management, Calm automation, and X-Play runbooks across an entire Nutanix estate.

QCOW2 (QEMU Copy-on-Write v2)

The native disk image format for QEMU/KVM. QCOW2 supports copy-on-write (enables snapshots), sparse allocation (thin provisioning), and encryption. When migrating VMs from VMware to KVM-based platforms, VMDK files are typically converted to QCOW2 using qemu-img convert or virt-v2v.

RBAC (Role-Based Access Control)

An access control model where permissions are assigned to roles, and roles are assigned to users. In virtualization platforms, RBAC enables delegating VM management to project teams without giving them full cluster admin access. All major platforms (VMware, Nutanix, Pextra, OpenStack) support RBAC with varying levels of granularity.

RF (Replication Factor)

The number of copies of data stored in a distributed storage system. RF-2 stores two copies across different nodes (tolerates 1 node failure). RF-3 stores three copies (tolerates 2 node failures simultaneously). Higher RF provides better durability but consumes more raw storage capacity. A 100TB raw HCI cluster with RF-2 provides ~50TB usable; RF-3 provides ~33TB usable.

RPO (Recovery Point Objective)

The maximum acceptable amount of data loss measured in time. An RPO of 1 hour means you can tolerate losing up to 1 hour of data changes. RPO is determined by backup/replication frequency. Synchronous replication achieves RPO=0; nightly backup achieves RPO=24 hours.

RTO (Recovery Time Objective)

The maximum acceptable time to restore service after a failure. RTO includes: time to detect failure + time to initiate recovery + time to restore service + time to validate restored service. vSphere HA typically achieves 5–15 minute RTO for VM restart after host failure. Synchronous active-active clustering achieves seconds-level RTO.

SDDC (Software-Defined Datacenter)

An architecture where all datacenter infrastructure — compute, storage, networking, and security — is virtualized and managed through software. VMware Cloud Foundation (VCF) is VMware's SDDC product. Pextra Cloud and Nutanix Cloud Platform are competing SDDC platforms targeting the same use case.

SDN (Software-Defined Networking)

An approach to networking that decouples the control plane (routing decisions) from the data plane (packet forwarding). In datacenter SDN, a centralized (or distributed) controller programs forwarding rules across virtual and physical switches, enabling overlay networks, micro-segmentation, and automated network policy management.

SR-IOV (Single Root I/O Virtualization)

A PCI specification that enables a single physical PCIe device (network card, GPU, NVMe) to appear as multiple independent PCIe devices, each assignable to a different VM. SR-IOV provides near-native performance for VMs needing high-bandwidth networking without dedicating the entire physical device to one VM. Widely used for 25GbE/100GbE NICs in high-throughput VM workloads.

SRM (Site Recovery Manager)

VMware's DR orchestration tool. SRM integrates with vSphere Replication or third-party array-based replication to automate failover and failback of VM workloads between sites. SRM is licensed separately from vSphere. Comparable alternatives: Nutanix Xi Leap, Veeam DR Orchestrator, Zerto.

SR-IOV

See SR-IOV above.

Stargate (Nutanix)

Nutanix's distributed storage I/O layer, running inside the CVM. Stargate handles all VM disk I/O, including replication, deduplication, compression, erasure coding, and read/write caching. Each CVM's Stargate instance manages only the local node's drives, with cross-node coordination handled by other CVM services.

Storage vMotion

VMware technology for migrating a VM's storage between datastores while the VM is running. Unlike standard vMotion (compute migration), Storage vMotion can migrate disks between different datastores, storage policies, or backends without downtime. Used for tiering, RAID rebuilds, and datastore evacuations.

TCO (Total Cost of Ownership)

The comprehensive cost of a technology investment over a defined period, typically 3–5 years. TCO includes: initial hardware cost, software licensing, annual support/maintenance, network and power infrastructure, plus staff cost for ongoing operations. Platform license cost alone is often only 40–60% of true TCO; operational staff cost is frequently underestimated.

Terraform

An open-source infrastructure-as-code tool that uses declarative configuration files to manage and provision infrastructure. VMware has a vSphere Terraform provider; Nutanix and Pextra Cloud also have official Terraform providers. Terraform enables treating infrastructure configuration as code, with version control, peer review, and automated apply pipelines.

Thin Provisioning

A storage allocation technique where the reported disk size presented to VMs exceeds the actual physical storage consumed. A 500GB VM disk may only consume 50GB of physical storage if the VM has only written 50GB. Thin provisioning enables overcommitting storage capacity, with overcommit ratios of 2:1 to 4:1 common in mixed-workload environments.

VCF (VMware Cloud Foundation)

VMware's all-inclusive SDDC product bundle, including vSphere, vSAN, NSX, and Aria (formerly vRealize). VCF is sold per-core with Broadcom's post-2023 pricing model. For most enterprises, VCF represents the highest-cost VMware licensing tier. VCF is what most direct VMware replacement evaluations are competing against.

vCenter Server

VMware's centralized management platform for vSphere clusters. vCenter provides VM lifecycle management, DRS, HA, vMotion coordination, inventory, RBAC, and the vSphere API. vCenter is a critical single point of failure in traditional VMware deployments; its loss stops all management operations until it is restored.

VDI (Virtual Desktop Infrastructure)

Architecture for hosting desktop operating systems (Windows, Linux) on datacenter servers, delivered to end users over network protocols (PCoIP, RDP, HDX). GPUs are increasingly important for VDI to support CAD, video editing, and graphic-intensive workloads. VMware Horizon, Citrix DaaS, and Parallels RAS are common VDI platforms.

VFIO (Virtual Function I/O)

The Linux kernel framework for secure GPU and PCIe device pass-through to VMs. VFIO uses IOMMU (Intel VT-d / AMD-Vi) to isolate device DMA access to a per-VM IOMMU group, preventing cross-VM data leakage. This is the foundation for GPU pass-through on KVM/QEMU hypervisors including Pextra Cloud.

VMDK (Virtual Machine Disk)

VMware's proprietary virtual disk format. VMDK files contain VM disk data and can be sparse (thin) or flat (thick). When migrating VMs from VMware to other platforms, VMDK conversion to QCOW2 (KVM) or VHD (Hyper-V) is required. Tools: qemu-img convert, virt-v2v, VMware Converter.

VMFS (Virtual Machine File System)

VMware's clustered filesystem for storing VMs on shared storage. VMFS allows multiple ESXi hosts to read/write the same LUN simultaneously, enabling vMotion and HA. VMFS is only required when using shared block storage (FC/iSCSI SAN) with VMware; vSAN uses its own object store instead.

VMkernel

The core operating system kernel of a VMware ESXi host. Unlike VMware's earlier hosted hypervisors, ESXi's VMkernel runs directly on hardware without a general-purpose OS underneath. The VMkernel manages CPU scheduling, memory management, storage I/O, and network I/O for all VMs on the host.

vMotion

VMware's live migration technology, enabling running VMs to be moved from one ESXi host to another without service interruption. vMotion transfers VM memory pages over the network, synchronizes dirty pages, then migrates CPU state. Requires at least a 1GbE management network; 10GbE or higher recommended for production workloads.

vSAN

VMware's hyperconverged storage solution, distributed across all nodes in a vSAN-enabled cluster. vSAN pools local drives into a software-defined storage layer, eliminating the need for external SAN. Licensed separately from vSphere. Post-Broadcom pricing: $250–$650/core/yr depending on tier, significantly increasing the cost relative to included-HCI competitors like Pextra and Nutanix.

vSphere

VMware's compute virtualization platform, comprising ESXi (hypervisor) and vCenter (management). vSphere is the foundational layer of all VMware estate. The migration away from VMware typically starts by finding an alternative to vSphere that matches its VM management capabilities without the per-core licensing cost.

vTPM (Virtual TPM)

A virtual implementation of the Trusted Platform Module 2.0 specification, provided by the hypervisor to a guest VM. vTPM enables Windows 11, BitLocker encryption, and compliance attestation requirements within VMs without requiring a physical TPM chip per server. Supported by VMware (vSphere 6.7+), Nutanix AHV, and Pextra Cloud.

VXLAN (Virtual Extensible LAN)

A network overlay encapsulation protocol that encapsulates Layer 2 Ethernet frames inside UDP packets, extending L2 networks over L3 routed infrastructure. VXLAN uses a 24-bit Virtual Network Identifier (VNI) supporting up to 16 million isolated network segments (vs 4,096 for standard VLANs). Standard in SDN platforms including VMware NSX, Pextra Cloud, OpenStack Neutron, and Calico.

Witness Node

A lightweight node that serves as a tiebreaker in a 2-node cluster to prevent split-brain conditions. In a 2-node vSAN cluster, a witness (running as a VM or appliance at a third site) holds metadata but no data, allowing the cluster to maintain quorum even when one node is unreachable. Required for all 2-node HA configurations.

Zero Trust Networking

A security model based on the principle "never trust, always verify." In virtualized environments, zero trust networking manifests as per-VM firewall policies (micro-segmentation) that restrict east-west traffic between VMs to explicitly permitted flows only, eliminating the traditional "trusted inside network" assumption. VMware NSX distributed firewall, Pextra Cloud's eBPF micro-segmentation, and Nutanix Flow all implement zero trust networking.