Publish Date: October 16, 2026
Executive Overview
Enterprise data center procurement and storage architecture have arrived at a contentious inflection point. Across global software-defined data centers running VMware Cloud Foundation (VCF), Chief Information Officers (CIOs), Chief Technology Officers (CTOs), and enterprise infrastructure architects face continuous pressure to curb hardware expenditures. Macroeconomic volatility, persistent server supply-chain friction, and semiconductor allocation shifts toward artificial intelligence accelerators have prompted infrastructure teams to scrutinize every line item on their server bills of materials (BOM). In an effort to minimize raw hardware acquisition costs, some organizations question why modern hyperconverged storage architectures reject legacy media. Specifically, infrastructure planners frequently ask whether low-cost rotational spinning media (magnetic HDDs) could be integrated into the storage tiers of the VMware vSAN Express Storage Architecture (vSAN ESA) to support secondary applications, archival data, or workloads with modest performance demands.
The definitive technical reality established by Broadcom is unambiguous: magnetic spinning disks do not work with, and are not supported in, vSAN Express Storage Architecture. Furthermore, vSAN ESA explicitly excludes legacy solid-state drives operating on Serial Advanced Technology Attachment (SATA) or Serial Attached SCSI (SAS) interfaces, as well as intermediate Tri-Mode RAID/HBA controllers. While legacy hybrid storage topologies—such as the vSAN Original Storage Architecture (vSAN OSA) hybrid model—successfully utilized high-performance NAND flash SSDs to buffer incoming write commands before staging them to magnetic spinning disks, that architectural paradigm was a product of early-2010s hardware constraints. Over the past decade, the rapid expansion of virtualized workload density, coupled with the systemic requirements of software-defined data availability and self-healing resilience algorithms, has rendered rotational media and legacy bus protocols a severe liability for modern storage fabrics.
This technical infrastructure advisory provides an architectural and economic evaluation of the deployment guidance published by Pete Koehler, Product Marketing Engineer in the VMware Cloud Foundation Division at Broadcom. Analyzing the microarchitectural physics, bus protocol inefficiencies, and queue depth constraints that separate legacy SAS/SATA spinning disks from native Non-Volatile Memory Express (NVMe) solid-state storage, this report diagnoses why vSAN ESA was engineered from the ground up exclusively for direct PCIe-attached NVMe devices. By evaluating the real-world operational impact on cluster self-healing and data evacuation timelines, alongside an objective total cost of ownership (TCO) model proving that advanced erasure coding (RAID-5/6) and global post-process deduplication achieve effective cost-per-usable-terabyte parity with spinning media, this advisory equips enterprise technology leadership with an authoritative framework to eliminate storage bottlenecks and future-proof their private cloud foundations.
Features
The architectural mandate restricting vSAN Express Storage Architecture exclusively to NVMe solid-state storage is rooted in a fundamental redesign of the storage data path. By stripping away decades of legacy hardware abstraction layers, vSAN ESA establishes a high-performance, single-tier storage processing engine designed to exploit the physical and protocol advantages of modern flash silicon.
The Physical Constraints of Rotational Media: Magnetic spinning disks are bound by rigid physical mechanics: magnetic platters rotating at 7,200, 10,000, or 15,000 revolutions per minute (RPM), accessed by physical mechanical actuator arms and magnetic read/write heads. These mechanical properties impose unavoidable seek times and rotational latencies. While historical enterprise storage systems utilized wide disk striping, short-stroking, and large DRAM caching layers to mitigate mechanical delays, these software workarounds delivered only marginal, incremental gains while failing to resolve the physical latency floor. When compared directly to modern NVMe solid-state storage, rotational media is up to 10,000 times slower in random throughput, delivers over 10,000 times fewer random IOPS, and introduces up to 2,000 times higher access latency.
Protocol and Queue Depth Bottlenecks (SATA and SAS): The bus protocols used by legacy media introduce severe processing bottlenecks. Both SATA and SAS were designed around the characteristics and limitations of magnetic platters, resulting in complex, legacy command sets that consume excessive host CPU cycles per I/O transaction:
Serial ATA (SATA): Operates as a half-duplex interface, meaning data transmission can only occur in a single direction at any given microsecond. SATA provides an extremely limited command buffer consisting of a single queue with a maximum depth of 32 commands (queue depth of 32).
Serial Attached SCSI (SAS): Provides full-duplex communication and expands queue capacity to a queue depth of 256 commands. However, it relies on legacy SCSI command parsing structures that generate high CPU overhead.
Architectural Mismatch with Flash: When NAND flash SSDs were retrofitted onto SAS and SATA interfaces in early all-flash arrays, the bus protocols became the primary system bottleneck, masking the native speed of flash silicon and under-delivering on solid-state performance expectations.
Elimination of Shared HBA Controller Funneling: In traditional SAS and SATA architectures, individual storage drives do not communicate directly with the host system. Instead, drives attach to a physical Host Bus Adapter (HBA) or RAID storage controller. In a high-density server blade, a single external controller is routinely responsible for arbitrating I/O across 8, 16, or 24 connected devices. This topology funnels all storage transactions through a single hardware choke point, introducing latency queues, driver contention, and complex failure domains. Furthermore, vSAN ESA explicitly prohibits the use of Tri-Mode controllers, which attempt to multiplex SAS, SATA, and NVMe drives across a shared controller chip, reintroducing the same funneling bottlenecks that ESA was engineered to eliminate.
Direct-Attach PCIe Bus Architecture in NVMe: NVMe abandons intermediary storage controllers entirely. NVMe solid-state devices connect directly to the server motherboard’s physical Peripheral Component Interconnect Express (PCIe) bus lanes. Under PCIe Gen 4 and Gen 5 specifications, a standard 4-lane ($x4$) NVMe solid-state device delivers up to 16 GB/s of bidirectional transfer throughput directly into host CPU root complexes. Every individual NVMe SSD incorporates its own dedicated, onboard hardware storage controller, decentralizing storage processing across physical drives and completely eradicating the centralized HBA bottleneck.
Massively Parallel Multi-Queue Scalability: Contrasting with SATA’s single queue of 32 commands and SAS’s queue of 256 commands, the NVMe specification supports up to 64,000 discrete queues, with each queue capable of holding up to 64,000 commands. This enables the ESXi hypervisor kernel to allocate dedicated, CPU-core-specific submission and completion queues directly to individual storage drives. Storage commands are processed locklessly across dozens of processor cores simultaneously, unlocking massive parallel execution that maintains flat, deterministic sub-millisecond latencies under dense multi-tenant workload saturation.
Lean, Flash-Optimized Command Primitives: Designed natively for solid-state non-volatile memory, NVMe utilizes a streamlined, low-overhead command set that discards decades of legacy SCSI/ATA register-mapping technical debt. Processing an I/O request over NVMe requires approximately one-third the CPU overhead per I/O compared to SATA, and between one-third to one-half the CPU overhead compared to SAS. This microarchitectural efficiency preserves substantial physical CPU socket capacity for user virtual machines and container runtimes.
Single-Tier Storage Architecture with 100% Capacity Contribution: Unlike vSAN Original Storage Architecture (OSA)—which mandated dividing physical drives into discrete disk groups consisting of a dedicated caching drive paired with multiple capacity drives—vSAN ESA operates as a unified, single-tier storage fabric. Every NVMe drive attached to an ESA host contributes 100% of its usable space to the cluster’s aggregate capacity pool. Cache-and-capacity disk groups are eliminated, lowering drive failure blast radiuses and removing the stranded hardware costs of dedicated caching media.
Formal Deprecation of Hybrid Configurations in Modern VCF Platforms: Reinforcing this architectural pivot, Broadcom officially announced the deprecation of hybrid (flash cache + spinning disk capacity) storage clusters in the vSAN Original Storage Architecture starting with the VMware Cloud Foundation 9.0 release. While all-flash configurations on vSAN OSA continue to be supported to facilitate phased enterprise migrations, net-new private cloud deployments are anchored to the all-NVMe baseline of vSAN ESA.
Benefits
Standardizing enterprise software-defined storage strictly on NVMe solid-state media within vSAN Express Storage Architecture yields transformative operational, architectural, and financial advantages over legacy multi-tier or hybrid storage approaches.
High-Velocity Data Evacuation and Platform Self-Healing: The performance of an enterprise storage fabric is directly linked to platform resilience and data availability. In a distributed software-defined storage system, vSAN continuously distributes data components across independent hosts and fault domains to maintain user-defined Failure to Tolerate (FTT) policies. When a physical drive fails, a network partition occurs, or an administrator places a host into maintenance mode for patching, vSAN must rapidly replicate and reconstruct data blocks across cluster nodes. Utilizing rotational media or slow SAS interfaces turns data rebuilds into days-long operational hazards, leaving data vulnerable to secondary hardware failures during extended rebuild windows. NVMe drives allow vSAN ESA to execute self-healing rebuilds, component scrubbing, and maintenance host evacuations in minutes rather than days, drastically compressing Mean Time to Repair (MTTR) and preserving continuous cluster availability.
Radical CPU Cycle Reallocation and Server TCO Optimization: Because NVMe utilizes a lean, hardware-aligned command set that operates locklessly over direct PCIe lanes, it consumes up to 66% fewer hypervisor CPU cycles per I/O transaction compared to SATA and SAS. In a multi-terabyte transactional environment, offloading storage arbitration overhead from physical CPU cores back to the application tier allows IT organizations to achieve higher virtual machine consolidation density per server blade. Enterprises extract greater application throughput per socket, maximizing per-core VMware Cloud Foundation subscription license investments and deferring server hardware additions.
Space-Efficiency Parity Dismantling the Raw Price Dilemma: While the upfront, raw “street price” of an enterprise NVMe SSD may appear up to 8x higher per gigabyte than an unmanaged magnetic spinning disk, this pricing gap collapses when factoring in the resilience mechanics of vSAN ESA:
Spinning Disk Constraints: Because spinning media lacks the random write performance required to process log-structured parity calculations, hybrid storage systems cannot support advanced RAID-5 or RAID-6 erasure coding without incurring severe write-penalty performance degradation. Consequently, spinning disk clusters are forced to rely on simple 3x mirroring ($FTT=2$) for mission-critical data, tripling the raw storage consumed. Furthermore, spinning disks cannot support inline data compression or post-process global deduplication.
NVMe-Enabled Space Efficiency: vSAN ESA implements adaptive, log-structured write paths that support high-performance RAID-5 and RAID-6 erasure coding with zero performance penalty. Storing data with RAID-6 resilience requires only 1.5x capacity overhead. When paired with vSAN ESA’s cluster-wide global deduplication and LZ4-based compression—delivering a standard 4:1 data reduction ratio—the effective capacity required to store one unit of data shrinks to just 0.375 units.
Financial Realization: The effective cost to store a unit of resilient enterprise data on high-performance NVMe reaches mathematical parity with spinning disks ($3 per effective unit), while delivering up to 10,000x faster I/O performance.
Eradication of Hardware Storage Controllers and Single-Point-of-Failure Bottlenecks: Removing physical RAID controllers and external SAS HBAs eliminates one of the most failure-prone, firmware-sensitive components in the enterprise server ecosystem. Systems engineers eliminate controller lockups, firmware driver mismatches, and HBA queue exhaustion events. Furthermore, eliminating external controllers reduces server bill-of-materials costs and simplifies hardware lifecycle auditing through vSphere Lifecycle Manager (vLCM).
Future-Proofing for Compressed Industry Lifecycle Demands: Enterprise computing demands are expanding toward memory-intensive real-time relational databases, microservices streaming, and private artificial intelligence inference pipelines. Attempting to support modern workloads with spinning disks creates immediate operational gridlock. Deploying vSAN ESA on direct-attach NVMe establishes an infrastructure foundation capable of supporting sub-millisecond AI agent loops, high-throughput vector retrieval, and dense transactional databases without requiring storage re-architecting.
Use Cases
Global organizations managing complex, latency-sensitive, and high-concurrency private clouds can evaluate the operational dividend of vSAN ESA’s NVMe architecture across diverse production scenarios.
High-Density Tier-1 Relational Database Fleet Consolidation: A multinational commercial banking conglomerate operates thousands of Microsoft SQL Server and Oracle Database instances across distributed data centers. Under legacy hybrid storage arrays, storage queue saturation during morning financial market openings triggered query latency spikes and database lock escalation. By modernizing onto VCF with vSAN Express Storage Architecture running certified Mixed-Use NVMe SSDs:
The bank eliminates physical RAID controllers, running all transactional database log and data files across direct PCIe Gen 4 NVMe lanes.
High-performance RAID-6 erasure coding reduces storage footprint by 50% compared to legacy 3x mirroring, while sub-millisecond write latencies prevent query stalls during peak market volumes.
When an unexpected host hardware failure occurs, vSAN ESA reconstructs distributed component replicas across cluster nodes in under twenty-five minutes—a workflow that previously required fourteen hours on spinning media—guaranteeing compliance with strict uptime mandates under the Digital Operational Resilience Act (DORA).
Healthcare Electronic Health Record (EHR) Platform Protection and Cluster Maintenance: A nationwide healthcare provider manages clinical imaging repositories, laboratory diagnostics, and electronic health records across regional hospital clusters. Operating under rigid HIPAA availability guidelines, the IT organization must execute monthly hypervisor security patching without impacting clinical portal responsiveness:
Under previous hybrid storage architectures, placing a host into maintenance mode required hours of pre-staging and created storage queue congestion that degraded diagnostic image retrieval in emergency rooms.
Utilizing vSAN ESA on NVMe storage, the clinical platform team executes rolling host evacuations in minutes. The 3.2 Tb/s backend network and direct PCIe storage pipelines stream data deltas across hosts without inducing hypervisor CPU stun.
The provider achieves continuous, uninterrupted 24/7 access to patient diagnostic records across hospital facilities while maintaining automated monthly patch schedules.
Sovereign Cloud Multi-Tenant Infrastructure-as-a-Service (IaaS) Delivery: A European sovereign cloud service provider delivers compliant private cloud hosting to municipal government agencies and regional law enforcement entities under strict NIS2 mandates:
To compete commercially against global public cloud hyperscalers, the provider must maximize storage capacity margins while guaranteeing rigid performance service-level agreements (SLAs).
Standardizing on vSAN ESA with high-density Read-Intensive NVMe SSDs, the provider enables cluster-wide post-process global deduplication across 64-host clusters, capturing a 4.2:1 data reduction ratio across tenant virtual machine disk images.
The provider achieves an effective cost-per-gigabyte that undercuts public hyperscaler storage tiers, delivering sovereign government tenants verifiable data isolation and line-rate storage performance without deploying complex external SAN switching fabrics.
Distributed Retail and Edge Logistics Hub Maintenance: A global digital retail enterprise operates automated fulfillment centers across hundreds of remote distribution hubs. Each facility runs localized computer-vision sorting algorithms, barcode scanning, and inventory tracking microservices on compact edge server nodes without on-site IT staff:
The retailer deploys vSAN ESA on NVMe storage across remote clusters, eliminating external storage arrays and fragile SAS cabling.
Because every NVMe device contributes directly to storage capacity without dedicated caching drives, edge nodes maximize physical drive bay utilization.
Self-healing background routines resolve transient drive errors automatically, ensuring that warehouse inventory pipelines process continuous shift operations without technician dispatches.
Alternatives
A comprehensive infrastructure assessment requires comparing the NVMe-native architecture of vSAN Express Storage Architecture against alternative private cloud storage and hardware deployment strategies.
Retaining Legacy Hybrid vSAN Original Storage Architecture (The Deprecated Path): Under this operational model, organizations attempt to extend the life of magnetic spinning disks by remaining on legacy vSAN OSA hybrid configurations (combining SAS/SATA flash caching drives with magnetic spinning disk capacity pools). While this approach allows enterprises to consume existing inventories of rotational media, it represents an obsolete, high-risk operational strategy. Broadcom formally announced the deprecation of hybrid vSAN OSA clusters in the VCF 9.0 release. Continuing to rely on hybrid clusters accumulates technical debt, forfeits access to modern platform capabilities (such as ESA snapshotting, native erasure coding, and advanced memory tiering), and exposes the organization to severe operational downtime when slow rebuilds fail to recover multi-terabyte drives before cascading hardware failures occur.
Traditional Three-Tier External Storage Area Networks (Fibre Channel / iSCSI SAN Arrays): In this conventional enterprise architecture, organizations reject hyperconverged storage convergence entirely, deploying external dedicated storage arrays connected to compute hypervisors via dedicated Fibre Channel or iSCSI switching fabrics. While high-end external all-flash SAN arrays provide mature data management features, they introduce massive architectural and financial complexity. External SANs mandate maintaining separate management consoles, paying recurring multi-vendor software maintenance contracts, managing complex multi-pathing drivers, and training specialized storage engineering teams. Furthermore, all storage traffic is forced across external switch fabrics and array controllers, reintroducing the centralized bottlenecking that vSAN ESA’s distributed NVMe architecture eliminates.
Public Cloud Managed Block and File Storage Services (AWS EBS / Azure Managed Disks / GCP Persistent Disk): Under this operational model, enterprises abandon on-premises private cloud storage modernization entirely, migrating workloads to public cloud hyperscalers and consuming cloud-managed virtual disks. While hyperscalers eliminate physical drive procurement and hardware maintenance, this strategy introduces long-term financial unpredictability. Public cloud storage billing enforces opaque, non-linear pricing models that charge aggressively for provisioned capacity, IOPS reservations, throughput bursts, snapshot retention, and network data egress. Over extended operational lifecycles, hosting tier-one database storage in the public cloud incurs operating expenditures that substantially exceed the amortized total cost of ownership of on-premises vSAN ESA NVMe private cloud infrastructure. Furthermore, public cloud multi-tenancy introduces data residency and compliance friction for regulated organizations.
Open-Source Software-Defined Storage on Commodity Media (Ceph / GlusterFS / DIY Storage Fabrics): In this architectural approach, internal platform engineering teams construct a custom software-defined storage cluster utilizing upstream open-source projects (such as Ceph) deployed across commodity bare-metal servers equipped with spinning disks and SATA SSDs. While this strategy avoids commercial software licensing fees, it shifts immense operational and financial risk onto internal engineering staff. Open-source storage fabrics lack native, kernel-level integration with the ESXi hypervisor, requiring complex iSCSI or NVMe-oF target mapping. In production, DIY open-source storage platforms routinely suffer from severe performance instability under tier-one relational database workloads, require specialized Linux storage kernel proficiencies, and lack certified enterprise Independent Software Vendor (ISV) support during critical storage corruption events.
Alternative Perspective
While the architectural arguments governing vSAN Express Storage Architecture’s exclusive reliance on NVMe solid-state storage are scientifically sound and operationally proven, an objective technical analysis reveals critical infrastructure prerequisites, supply-chain considerations, and transition challenges that enterprise platform leadership must evaluate prior to enterprise deployment.
A primary technical consideration centers on the physical hardware and motherboard prerequisites demanded by direct-attach NVMe storage. Because NVMe drives interface directly with physical PCIe lanes, deploying vSAN ESA requires modern server platforms equipped with sufficient PCIe lane allocations, native backplanes (such as U.2, U.3, or E3.S form factors), and certified PCIe root complex switches. Organizations operating older, legacy server blades designed with backplanes hard-wired exclusively to SAS/SATA RAID controllers cannot simply swap spinning disks for NVMe drives; adopting vSAN ESA requires a coordinated physical server hardware refresh. Technology leadership must account for these server hardware capital expenditures when planning their private cloud modernization budgets.
Furthermore, platform architects must recognize that while NVMe delivers massive raw throughput and low latency, unlocking the full performance of vSAN ESA introduces strict physical networking underlay dependencies. A distributed storage fabric that writes and rebuilds data across hosts at NVMe line rate requires an equally capable network transport tier. Broadcom’s validated design guidelines mandate a minimum of 25GbE networking for vSAN ESA, with 40GbE or 100GbE strongly recommended for high-density environments. If an enterprise attempts to deploy vSAN ESA atop an aging 10GbE physical network fabric, the network switch uplinks will immediately become the primary system bottleneck, throttling storage throughput, inflating replication latencies, and blunting the operational return of NVMe media.
Finally, enterprise leadership must address the procurement governance and drive-class selection required within NVMe storage tiers. The term “NVMe” describes a communication interface, not a uniform performance or endurance tier. NVMe solid-state drives encompass a broad spectrum of hardware classes, ranging from low-cost, low-endurance Read-Intensive (1 DWPD) drives utilizing 7% over-provisioning to high-performance, high-endurance Mixed-Use (3 DWPD) drives utilizing 28% over-provisioning. If an organization blindly procures entry-level Read-Intensive NVMe drives for write-heavy database or advanced memory-tiering workloads, the drives can experience controller garbage-collection performance cliffs that degrade storage responsiveness. Platform engineering teams must enforce rigid procurement baselines aligned strictly with the VMware Compatibility Guide, ensuring that drive endurance classes, throughput profiles, and firmware baselines match their specific workload profiles.
Final Thoughts
The debate regarding whether magnetic spinning disks have a place in modern enterprise private cloud architectures has been settled by the physics of semiconductor computing and the mathematical requirements of distributed software-defined storage. In an era where enterprise applications demand deterministic sub-millisecond execution and private cloud platforms must maintain continuous self-healing data resilience, rotational media and legacy SAS/SATA interfaces represent an obsolete, fragile architectural compromise. Broadcom’s strategic mandate restricting VMware vSAN Express Storage Architecture exclusively to direct-attach NVMe solid-state devices is an empirical engineering decision that eliminates hardware choke points and unlocks the full potential of modern flash silicon.
By abandoning the shared HBA controllers, half-duplex single queues, and mechanical latencies of legacy architectures, vSAN ESA transforms the private cloud storage fabric. When factoring in the compounding efficiencies of native RAID-5/6 erasure coding, inline compression, and cluster-wide global deduplication, NVMe solid-state storage achieves effective cost parity with spinning media while delivering up to 10,000 times higher random I/O performance and compressing cluster rebuild windows from days down to minutes.
Enterprise technology leadership should take immediate, decisive operational steps: audit existing virtualization estates to identify legacy hybrid vSAN OSA clusters slated for deprecation, revise upcoming server procurement specifications to mandate direct-attach PCIe NVMe backplanes (U.2/U.3/E3.S) and 25GbE/100GbE physical switching fabrics, and model storage capacity requirements based on vSAN ESA’s data reduction algorithms rather than raw drive street prices. By anchoring enterprise private cloud infrastructure to a validated, NVMe-native storage foundation, organizations can eliminate storage-induced downtime, maximize compute consolidation density, and build an impenetrable operational fabric for continuous enterprise innovation.