
The Storage Policy-Based Management (SPBM) directly inside vCenter is easily the best part for me.
Coming into AI and ML pipelines from a standard automation background, I expected storage provisioning to be a bit of a bottleneck with lots of back-and-forth tickets. But with vSAN, I can just apply a policy based on what the job needs—like assigning low latency and better striping for fast-access scratch disks when preprocessing training datasets, or lower performance tiers for basic archive logs—right when deploying the VMs or nodes.
Not having to worry about physical LUNs or separate external SAN arrays cuts out so much manual infra coordination. It basically turns storage into just another software setting we can handle directly within the familiar VMware UI, which lets us spin up new experimental pipelines and agent test environments much faster. Review collected by and hosted on G2.com.
Mainly the licensing costs and hardware pickiness.
The Hardware Compatibility List (HCL) is very strict—if an SSD model or controller firmware is slightly off, you will see warnings. Also, during heavy uncompressed data ingest for ML batches, if the cache tier gets saturated, write latency spikes.
On top of that, licensing feels pricey, and when a disk group acts up or data resync kicks in, figuring out the root cause from the vCenter UI isn't very intuitive without digging through logs. Review collected by and hosted on G2.com.


