6.1 vSAN Core Architecture: OSA vs Express Storage Architecture (ESA)

6.1 vSAN Core Architecture: OSA vs Express Storage Architecture (ESA)

Traditional enterprise datacenters have historically separated compute resources from storage infrastructure, relying on dedicated Storage Area Networks (SAN) or Network-Attached Storage (NAS) arrays connected via Fibre Channel or iSCSI fabrics. While functional, this bifurcated model introduces operational complexity, storage controller bottlenecks, and complex capacity forecasting. VMware vSAN transforms this model by delivering software-defined, hyperconverged storage (HCI) built directly into the VMware ESXi hypervisor.

---

The Hyperconverged Storage Model & Distributed Object Architecture

Unlike legacy Virtual Storage Appliances (VSAs) that execute as virtual machines running a third-party storage operating system, vSAN is embedded directly into the ESXi VMkernel at the I/O data path. This native kernel integration delivers two primary architectural advantages: it eliminates the context-switching latency overhead inherent to user-space virtual machines, and it allows storage operations to access CPU and memory scheduling primitives with direct integration into the hypervisor storage path.

vSAN aggregates local, direct-attached storage devices (NVMe, SAS, SATA solid-state drives, or magnetic hard disks) across all participating ESXi hosts in a cluster into a single, shared distributed object datastore known as vsanDatastore. Every host in the cluster—whether it contributes local storage or operates as a compute-only node—can access this shared datastore subject to network and placement conditions.

The Object-Based Storage Paradigm

Traditional file systems organize storage into blocks and tracks. vSAN manages data as discrete storage objects. An object is a distinct logical volume that represents a specific virtual machine entity, backed by one or more physical components distributed across cluster nodes according to assigned Storage Policy-Based Management (SPBM) rules:

• VM Home Namespace: The directory containing VM configuration files (.vmx), log files, and descriptor files.
• VMDK (Virtual Disk): The virtual hard disk containing the guest operating system and user data.
• VM Swap: Created when a virtual machine powers on, sized to match unreserved memory allocations.
• Snapshot Delta Disks: Created when point-in-time virtual machine snapshots are generated.
• Memory State: Preserves guest physical memory when taking a snapshot with memory or suspending a VM.

At the hypervisor layer, the Distributed Object Manager (DOM) orchestrates I/O distribution, replication, and quorum across hosts, while the Local Log-Structured Object Manager (LSOM) interfaces directly with local physical storage devices.

---

vSAN Original Storage Architecture (OSA)

The vSAN Original Storage Architecture (OSA) represents the foundational two-tier storage model utilized since the inception of vSAN. In OSA, physical storage devices inside each ESXi host are partitioned into administrative units called Disk Groups.

Sizing Rules & Device Ratios

Every OSA disk group adheres to strict physical configuration rules:

1. Fixed Cache Ratio: Exactly one dedicated flash-based cache device per disk group. You cannot configure multiple cache drives within a single disk group.
2. Capacity Tier Sizing: Between one and seven capacity devices per disk group.
3. Host Limits: An ESXi host can support a maximum of five disk groups, permitting up to 5 cache devices and 35 capacity devices per physical host.

Hybrid vs. All-Flash Operational Modes

OSA operates in two distinct hardware configurations:

• Hybrid Configuration: Combines flash-based solid-state drives in the cache tier with magnetic spinning disks (HDDs) in the capacity tier. The cache device is split logically: 70% of the cache space is allocated as a read cache to store frequently accessed blocks (accelerating read cache hits), while 30% is allocated as a write buffer to absorb incoming write bursts before destaging to spinning disks.
• All-Flash Configuration: Utilizes flash SSDs (SAS, SATA, or NVMe) in both the cache tier and capacity tier. Because flash capacity devices provide sub-millisecond random read latencies, a dedicated read cache is unnecessary. Therefore, 100% of the cache device is dedicated as a write buffer. Incoming writes are rapidly committed to the write buffer and later destaged sequentially to capacity flash drives.

Cache and Capacity Planning

In OSA, size cache and capacity from the supported ReadyNode/device combination and workload demand. The logical write-buffer limit is release-specific; drive size alone does not prove usable buffer capacity. Validate controller mode, firmware, drivers and device endurance against the compatibility guide. A generic cache percentage or drive-writes-per-day number is not a substitute for a certified design.

The Failure Blast Radius in OSA

The primary architectural liability of OSA is its blast radius. Because the single cache device serves as the exclusive entry gateway for all read and write I/O within the disk group, a physical failure of that single cache drive causes the entire disk group to go offline. All capacity drives attached to that disk group immediately become unavailable, forcing vSAN to initiate component rebuilds across the remaining cluster nodes.

---

vSAN Express Storage Architecture (ESA)

Introduced in vSphere 8 and enhanced in vSphere 9, the vSAN Express Storage Architecture (ESA) is an optional, alternative architecture engineered from the ground up for modern enterprise hardware. High-performance PCIe NVMe solid-state drives, high-core-count processors, and 25 GbE / 100 GbE networking allow storage to be handled without the legacy constraints of disk groups.

The Single-Tier NVMe Storage Pool

ESA completely eliminates the two-tier disk group hierarchy and dedicated cache drives. Instead, ESA introduces a single-tier storage pool:

• Uniform Drive Participation: Every drive claimed by ESA is a certified NVMe SSD that contributes simultaneously to both write performance and persistent capacity storage.
• Elimination of the Cache Blast Radius: Because there are no dedicated cache drives, the failure of an individual NVMe drive impacts only the specific components stored on that device. It never causes sibling drives to go offline, drastically reducing the failure domain within the host.

The Log-Structured File System (vSAN LFS)

At the heart of ESA is the vSAN Log-Structured File System (vSAN LFS). In traditional storage architectures, writing small, random data blocks incurs significant write amplification and CPU overhead. vSAN LFS ingests incoming writes rapidly into an in-memory and NVMe-backed write log. The LFS batches and packages writes into full stripes, writing them sequentially to NVMe capacity. This design minimizes write amplification, extends SSD drive longevity, and frees CPU cycles.

Adaptive RAID-5/6 Erasure Coding with RAID-1 Write Performance

In the Original Storage Architecture, selecting RAID-5 or RAID-6 erasure coding saved disk capacity but imposed a severe performance penalty. Every write operation required a "read-modify-write" cycle—reading existing parity and data blocks, calculating new parity, and writing back the changes—creating noticeable write latency.

vSAN ESA eliminates the erasure coding write penalty. By leveraging the Log-Structured File System, ESA writes full stripes without performing read-modify-write cycles, achieving RAID-1 write performance parity while maintaining the capacity savings of RAID-5 or RAID-6.

Furthermore, ESA introduces Adaptive RAID-5:

• Small Clusters (3 to 5 Hosts): Automatically configures a 2+1 parity scheme (2 data segments + 1 parity segment), requiring a minimum of 3 hosts and introducing a modest 1.5x capacity multiplier (33% parity overhead).
• Standard Clusters (6 or More Hosts): Automatically transitions to a 4+1 parity scheme (4 data segments + 1 parity segment), requiring a minimum of 5 hosts and delivering a 1.25x capacity multiplier (only 20% parity overhead).

Snapshot Behavior

ESA's native snapshot implementation differs from traditional redo-log chains and is designed to reduce snapshot overhead. That does not make snapshots independent backups or guarantee zero latency during every operation. Validate backup-product interoperability, policy, retention and recovery. Keep an independent recovery copy for failures outside the protected storage cluster.

Comprehensive Comparison: OSA vs. ESA

Official Reading

Use the official 2V0-16.25 objective guide for scope and Broadcom technical documentation for the installed 9.0 component. This independent lesson was reviewed October 18, 2026; actual deployment support depends on the applicable version and configuration.

Source Notes

See ESA network profiles and adaptive RAID-5 design. Checked October 18, 2026.

Review Takeaways

• OSA uses disk groups with a dedicated cache device and capacity devices.
• ESA uses a storage pool per contributing host and certified NVMe devices without dedicated cache disk groups.
• A cache-device failure can make an OSA disk group unavailable; ESA has a different per-device failure boundary.
• ESA network requirements depend on the supported ReadyNode profile, including supported 10 GbE entry configurations.