When VMware announced vSAN 8 Express Storage Architecture (ESA), the marketing language was predictably breathless. Having now run ESA clusters in production for several months, I can separate the genuine architectural advances from the hype.
The Log-Structured Engine
The most fundamental change in ESA is the move to a log-structured storage engine. Where vSAN OSA used a magnetic journal + SSD cache tier + capacity tier design, ESA writes all data sequentially to a log before compacting it asynchronously.
This matters for two reasons:
- Write amplification drops dramatically. Random writes become sequential log appends. In our testing with a OLTP-style workload, write amplification went from ~4x (OSA) to ~1.3x (ESA).
- The cache tier disappears. ESA requires NVMe-only storage but uses all drives as a single-tier capacity pool. Your all-NVMe investment now goes entirely to capacity rather than splitting 25%/75% between cache and capacity.
Hardware Requirements
ESA's NVMe-only requirement is a hard gate. The officially supported TLC NVMe drives are listed in the VMware Compatibility Guide, and the list is intentionally narrow — ESA relies on specific NVMe command sets for its compaction I/O patterns.
Minimum per host for ESA:
- 4x NVMe drives (each ≥ 1.6 TB)
- 512 GB RAM (recommended)
- 25 GbE networking (100 GbE for all-flash performance tiers)
If you're speccing new hardware, go 100 GbE from the start. Storage networking headroom disappears faster than compute headroom in high-density NVMe environments.
RAID-5/6 Performance
ESA's erasure coding (RAID-5 and RAID-6) performance is genuinely competitive with RAID-1 on OSA, because the log-structured engine batches erasure coding operations during compaction rather than computing parity on every write. For stretched cluster deployments needing RAID-6 across sites, this is a significant operational improvement.
What ESA Doesn't Change
- Operational model: Cluster sizing, fault domains, and witness traffic patterns remain the same.
- vSphere integration: ESA surfaces as standard datastores to the hypervisor. VM configurations don't change.
- Encryption: Data-at-rest encryption via vCenter Key Provider works identically.
Should You Migrate?
ESA is a new cluster format — you can't migrate OSA clusters in-place. Factor in fresh hardware procurement and a proper data migration project. For greenfield deployments on NVMe hardware, ESA should be your default. For existing OSA clusters that are performing well, hold your position until the next hardware refresh cycle.