VMware vSAN OSA and ESA Algorithm overview

VMware vSAN's Original Storage Architecture (OSA) uses a two-tier model with disk groups, while the newer Express Storage Architecture (ESA), introduced in vSAN 8, utilizes a single-tier model with a log-structured file system optimized for modern, high-performance NVMe storage. 

Original Storage Architecture (OSA) Overview

The OSA was designed to be compatible with a wide range of hardware, including hybrid (flash cache, spinning disk capacity) and all-flash configurations. 

Two-Tier Model: It separates storage devices into distinct cache and capacity tiers organized into disk groups.

Disk Groups: Each disk group has one flash device for caching and buffering writes (and reads in hybrid configurations) and one to seven devices for persistent capacity.

Data Flow & Processing:

Writes are first sent to the high-performance cache tier (write buffer) and acknowledged quickly to the VM.

Data is then "destaged" to the capacity tier in a process that involves a read-modify-write step for space-efficient policies like RAID-5/6 erasure coding.

Data services like deduplication and compression operate within the scope of a single disk group.

Failure Domain: A failure of the cache device affects the entire disk group, leading to a larger data resynchronization effort and impact on performance. 

Express Storage Architecture (ESA) Overview

The ESA is a re-architecture designed to exploit the full potential of high-performance NVMe drives, faster CPUs, and high-speed networking (10 GbE or higher recommended). 

Single-Tier Model (Storage Pools): The concept of disk groups is eliminated. All eligible NVMe devices in a host are pooled into a single storage pool, where every device contributes to both performance and capacity.

Data Flow & Processing (Log-Structured File System):

ESA uses a new, patented log-structured file system (vSAN LFS) and an optimized log-structured object manager.

It coalesces many small, incoming writes in memory before persisting them to disk in a highly efficient, full-stripe write operation. This eliminates the need for the read-modify-write penalty of OSA's erasure coding.

Data services (compression, encryption, and checksums) are performed higher in the I/O stack (closer to the VM) and only once, reducing CPU and network overhead. Compression in ESA is more granular and effective than in OSA.

Deduplication in ESA is a post-process activity that operates globally across the entire cluster, offering much higher efficiency compared to OSA's per-disk-group deduplication.

Failure Domain: The failure domain is reduced to a single discrete disk, significantly minimizing the amount of data and time needed for resynchronization compared to an entire disk group failure in OSA.

Snapshots: ESA introduces a new native snapshot engine with minimal performance impact, improving the performance of backup solutions and snapshot operations. 

In summary, the ESA offers a simpler management experience, higher performance, and better efficiency through a modernized, hardware-optimized data path, making it the preferred choice for new deployments on compatible hardware. Customers with existing, older hardware can continue to use and upgrade the OSA. 

Cheers !!