OEM Evaluation
Enterprise Storage OEM Comparison
| OEM | Strengths | Best For |
|---|---|---|
| Pure Storage (FlashArray/FlashBlade) | Simplicity, Evergreen upgrades, strong AI/ML performance | All-flash, AI workloads, organizations valuing simplicity |
| Dell (PowerStore/PowerScale) | Broad portfolio, strong support, VMware integration | Mixed workloads, VMware environments, large enterprises |
| NetApp (AFF/StorageGRID) | ONTAP data management, strong NAS, object storage | NAS-heavy environments, hybrid cloud, compliance |
| HPE (Alletra/Primera) | GreenLake as-a-service, strong support | Organizations preferring OpEx model, HPE compute environments |
| IBM (FlashSystem) | Strong mainframe integration, Spectrum Virtualize | IBM-centric environments, mainframe workloads |
| Vast Data | Universal storage, AI-optimized, NVMe/QLC | AI/ML workloads, large unstructured data |
Sizing and Capacity Planning
Storage sizing must account for effective capacity (after data reduction), not just raw capacity. All-flash arrays achieve 2–5:1 data reduction ratios for typical enterprise workloads, but ratios vary significantly by workload type. Databases with random data achieve lower ratios (1.5–2:1); virtual machine images achieve higher ratios (3–5:1).
Verify data reduction ratios with your data
Current data footprint
Total data currently stored, including all copies, snapshots, and replicas.
Data growth rate
Historical growth rate plus projected growth from new workloads. AI datasets grow faster than traditional enterprise data.
Snapshot overhead
Snapshot space consumption depends on change rate. High-change workloads (databases) consume more snapshot space than low-change workloads.
Performance headroom
Storage arrays perform best at 70–80% capacity utilization. Size for 80% utilization at the end of the planned lifecycle.
RFP Requirements
Performance at workload level
Require IOPS, throughput, and latency specifications at the specific workload mix, not peak specifications under ideal conditions.
Effective capacity with verified data reduction
Require effective capacity based on proof-of-concept testing with the organization's actual data, not vendor-claimed ratios.
Non-disruptive upgrade path
Require documentation of the upgrade path for controller, software, and capacity additions, and whether each upgrade requires data migration.
Data migration methodology
Require documentation of the data migration methodology for end-of-life replacement, including estimated downtime and migration duration.
Encryption capabilities
Require documentation of encryption at rest capabilities: algorithm, key management, and compliance certifications (FIPS 140-2).
Proof-of-Concept Testing
Proof-of-concept testing with actual production workloads is the only reliable way to verify storage performance and data reduction claims. A PoC should run for at least 30 days: long enough to capture the full range of workload patterns, including peak periods.
PoC success criteria should be defined before the test begins: not after. Criteria should include: IOPS and latency at peak load, effective capacity after data reduction, and management platform usability. Vendors who resist PoC testing or propose short PoC periods are not confident in their product's performance with your workloads.
Contract Terms
Evergreen upgrade commitment
Vendors who commit to non-disruptive controller upgrades eliminate the data migration cost at end of controller life. Verify the specific terms, not all "evergreen" programs are equivalent.
Support response time
Mission-critical storage requires 4-hour or better on-site response. Verify that the vendor has local support resources, not just a remote support center.
Capacity guarantee
Some vendors offer capacity guarantees, committing to a minimum effective capacity after data reduction. Require this guarantee in writing with specific remedies if not met.
Data migration assistance
Require the vendor to provide data migration assistance at end of life: including tools, methodology, and professional services support.