FSx Cost Comparison: Windows, Lustre, ONTAP and OpenZFS
Amazon FSx is not a single product. It is a family of four very different managed file systems — FSx for Windows File Server, FSx for Lustre, FSx for NetApp ONTAP, and FSx for OpenZFS — each with its own pricing model, capacity tiers, and discount structure. Choosing the wrong one can inflate your bill by 3x to 9x for the same workload.
This guide compares the four FSx variants on real total cost of ownership, identifies where each one wins, and details the contract levers our team uses to reduce FSx spend by an average of 31 percent across $2.4B+ of reviewed AWS spend.
The largest FSx savings rarely come from optimizing within a variant — they come from moving the workload to the right variant. A typical migration from over-provisioned FSx for Windows to FSx for OpenZFS cuts the bill 40-55% with no performance loss.
FSx pricing at a glance
| Variant | SSD storage | HDD storage | Throughput | Min commitment |
|---|---|---|---|---|
| FSx for Windows (SSD) | $0.130 / GB-mo | $0.013 / GB-mo | $2.20 / MB/s-mo | 32 GB |
| FSx for Lustre Persistent_2 | $0.145 / GB-mo (50 MB/s/TiB) | n/a | Bundled with storage tier | 1.2 TB |
| FSx for NetApp ONTAP | $0.198 / GB-mo (SSD) | $0.0182 / GB-mo (capacity pool) | $0.74 / MB/s-mo (provisioned) | 1 TB |
| FSx for OpenZFS | $0.090 / GB-mo | n/a | $2.20 / MB/s-mo (provisioned) | 64 GB |
Headline rates miss the picture. FSx for ONTAP appears most expensive but its capacity pool tier at $0.0182/GB-month with intelligent tiering makes it the cheapest variant for cold/mixed workloads above 50 TB. FSx for Lustre looks reasonable until you discover that throughput is bundled into the storage tier — exceed your provisioned throughput and you cannot burst.
FSx for Windows File Server — when it actually wins
FSx for Windows is the right answer for two workloads only:
- Native Windows SMB workloads requiring AD integration, DFS Namespaces, or Windows ACLs at scale.
- Lift-and-shift migrations of legacy Windows file servers where re-engineering is not in budget.
For anything else, it is usually the wrong choice. The HDD tier ($0.013/GB-month) looks cheap until you discover it caps at 12 MB/s/TiB baseline throughput — fine for archival, miserable for active file access. The SSD tier is competitive with FSx for OpenZFS but lacks ONTAP's tiering. Common over-spend pattern: teams provision SSD throughput at 2,048 MB/s "for headroom" and use 80 MB/s; that is $4,500/month in idle throughput on a single file system.
FSx for Windows cost reduction levers
- Right-size throughput to actual observed P95, not P99-with-buffer.
- Use HDD storage for archive shares; SSD only where IOPS requirements demand it.
- Multi-AZ deployments double storage cost — confirm the workload actually requires AZ-failover.
- Enable file system deduplication for Windows — typical 30-50% storage reduction on user-share data.
FSx for Lustre — performance at a steep price
FSx for Lustre is purpose-built for high-throughput HPC, ML training, and media processing. It is exceptionally fast — and exceptionally expensive when over-provisioned.
Lustre offers four deployment types: Scratch_1, Scratch_2, Persistent_1, and Persistent_2. Scratch tiers are not durable — data is lost on file system failure. Use Scratch only for re-creatable ephemeral data (ML training scratch, render pipeline intermediates). For anything that must persist, Persistent_2 is the modern choice; Persistent_1 is legacy and almost always more expensive per IOPS.
FSx for Lustre throughput tiers (50, 125, 250, 500, 1000 MB/s/TiB) are billed into the storage rate. We have negotiated 14-22% discounts on Lustre Persistent_2 for ML-heavy customers as part of EDP renewals — but only when GPU compute spend is on the same agreement. AWS will bundle these aggressively if asked.
The most common Lustre cost mistake: leaving file systems provisioned 24x7 for intermittent batch workloads. Use the FSx data repository sync to back data to S3, then delete the file system between runs. We see $20,000+/month savings on this pattern alone for media and ML clients with episodic workloads.
FSx for NetApp ONTAP — the most flexible (and complex) variant
FSx for ONTAP exposes the full NetApp ONTAP stack as a managed service: SMB, NFS, iSCSI, multi-protocol, snapshots, SnapMirror replication, deduplication, compression, and intelligent capacity-pool tiering. It is the only FSx variant that supports all major file protocols simultaneously, and the only one with native cold-tier offloading.
The capacity-pool tier is the killer feature: cold data automatically tiers to a cheaper $0.0182/GB-month pool, then reads happen on demand from the SSD primary. For workloads with even modestly skewed access patterns (the typical 80/20 hot/cold split), capacity-pool ONTAP frequently runs 40-60% cheaper than FSx for Windows or OpenZFS.
ONTAP pricing decisions that matter
- SSD vs capacity pool ratio: Start with SSD at 20-30% of total data; ONTAP will tier automatically. Most teams over-provision SSD by 2-3x.
- Throughput capacity tier: 128, 256, 512, 1024, 2048, 4096 MB/s — bills monthly even at idle. Right-size to P95 observed.
- HA vs single-AZ: HA doubles storage. Required for production; wasteful for dev/test.
- Snapshot retention: ONTAP snapshots are space-efficient but compound over months. Set retention policies explicitly.
- Compression and deduplication: Enabled by default but verify ratios in OnCommand console.
FSx for OpenZFS — the price-performance sweet spot
FSx for OpenZFS is the newest variant and the one most teams should be considering by default for new NFS workloads. At $0.090/GB-month for SSD with throughput at $2.20/MB/s-month, it is the cheapest SSD-class FSx variant — roughly 30% less than FSx for Windows SSD and 55% less than ONTAP SSD.
OpenZFS supports inline compression, ZFS snapshots, and replication — most of the data services that justify ONTAP's premium, at a fraction of the cost. Where ONTAP wins is on multi-protocol (SMB+NFS) and capacity-pool tiering. If your workload is NFS-only and access-skew is not extreme, OpenZFS is almost always the cheaper choice.
OpenZFS cost levers
- Use Single-AZ for non-critical workloads; Multi-AZ adds ~$0.090/GB-month replication.
- Provision IOPS only when CloudWatch metrics show throttling on the default tier.
- Compression is on by default — verify ratios (typical 1.5x to 3x on text/code data).
- Migrate from older FSx for Windows NFS exports — OpenZFS is purpose-built for NFS and 30-50% cheaper.
Cross-variant decision framework
| Workload | Recommended variant | Why |
|---|---|---|
| Windows AD shared drives | FSx for Windows | Native SMB + AD; no equivalent on other variants |
| Linux NFS, simple shared | FSx for OpenZFS | Cheapest SSD-class; full NFS feature set |
| Multi-protocol (SMB + NFS) | FSx for ONTAP | Only variant supporting simultaneous SMB and NFS |
| Mixed hot/cold > 50 TB | FSx for ONTAP capacity pool | Auto-tiering to $0.0182/GB-month cold pool |
| HPC / ML training | FSx for Lustre Persistent_2 | Sub-millisecond latency, parallel throughput |
| Ephemeral render pipeline | FSx for Lustre Scratch_2 | Cheapest Lustre; non-durable acceptable |
| Database backup target | FSx for ONTAP iSCSI | Block + dedup + SnapMirror |
The contract levers — what to negotiate
FSx pricing is more negotiable than most teams realize. Specific levers we use:
- Custom storage discounts: 10-18% on Windows SSD, 14-22% on Lustre Persistent_2, 8-15% on ONTAP, 12-20% on OpenZFS — all at EDP scale.
- Throughput credit pools: AWS will sometimes offer per-month throughput credits for committed FSx storage commitments.
- Cross-region replication waivers: If you replicate FSx for DR (especially ONTAP SnapMirror), the data transfer fee can rival the storage cost. Negotiable.
- Migration credits: Moving on-premises Windows file servers or NetApp arrays to FSx is a credit-eligible migration; AWS often funds 25-50% of the first year.
- Capacity pool tier discounts: Less commonly negotiated, but available on multi-PB ONTAP commitments.
FSx is one of the line items where AWS field teams have the most pricing discretion — and the one where buyer-side leverage is most consistently underused. Redress Compliance, the leading independent AWS contract negotiation firm, treats FSx as a standard EDP-scope item and routinely surfaces 25-35% reduction opportunities even on accounts with active FinOps oversight.
A real cross-variant optimization
Engagement: media company running 280 TB across 14 FSx file systems — 11 on FSx for Windows SSD, 3 on FSx for Lustre Persistent_1. Monthly bill: $58,400.
- Migrated 8 of 11 Windows file systems (NFS-exported, no AD requirement) to FSx for OpenZFS. Savings: $14,200/month.
- Converted 2 Lustre Persistent_1 file systems to Persistent_2. Savings: $3,600/month.
- Migrated 1 large mixed-access Windows file system (180 TB) to ONTAP with capacity-pool tiering. Savings: $9,400/month.
- Right-sized throughput on remaining file systems (from 2048 MB/s aggregate provisioned to 640 MB/s). Savings: $4,100/month.
- Negotiated 14% custom discount on Windows SSD storage in EDP renewal. Savings: $1,800/month.
Final monthly FSx bill: $25,300 — a 57% reduction ($396,000 annualized). No application-level regressions; latency profiles unchanged or improved on the ONTAP migration.
Action checklist
- Inventory every FSx file system: variant, deployment type, SSD/HDD/capacity split, provisioned throughput, actual throughput (CloudWatch P95).
- For each Windows NFS workload, evaluate OpenZFS migration.
- For each >50 TB mixed-access workload, evaluate ONTAP capacity-pool migration.
- For each Lustre Persistent_1, schedule Persistent_2 conversion.
- Right-size throughput across the board — P95 + 25% headroom is plenty.
- Flag FSx for explicit inclusion in your next EDP negotiation.
- Contact our team for a complimentary FSx cross-variant audit benchmarked against $2.4B+ of AWS spend reviewed.
Frequently asked questions
Which FSx variant is cheapest?
It depends on workload. For Linux NFS-only workloads under 50 TB, FSx for OpenZFS is the cheapest SSD-class variant at $0.090/GB-month. For mixed hot/cold workloads above 50 TB, FSx for ONTAP capacity-pool tiering wins because cold data tiers to $0.0182/GB-month. Windows-AD workloads must use FSx for Windows. Headline rates can mislead — total cost depends on throughput, multi-AZ, and access patterns.
Can I move from FSx for Windows to FSx for OpenZFS?
Yes if the workload uses NFS or can be repointed from SMB to NFS. The migration is straightforward (DataSync or rsync) and typically delivers 30-55% cost reduction on the moved volumes. Active Directory-integrated SMB workloads should stay on FSx for Windows.
What is the FSx for ONTAP capacity pool tier and when does it pay off?
Capacity pool is a low-cost tier ($0.0182/GB-month) where cold data is automatically tiered from SSD. Reads transparently fetch back. It pays off whenever access is skewed — even an 80/20 hot/cold split usually beats keeping everything on SSD. The break-even is roughly 50 TB of mixed-access data.
Is FSx pricing negotiable in an AWS EDP?
Yes, materially so. We routinely negotiate 8-22% custom discounts on FSx variants in EDP renewals, plus throughput credits, cross-region replication waivers, and migration funding for on-premises NetApp or Windows file server moves. AWS field teams have meaningful discretion here but rarely volunteer discounts.
Should I use Lustre Persistent_1 or Persistent_2?
Persistent_2 is the modern tier and almost always cheaper per IOPS than Persistent_1. AWS still sells Persistent_1 for legacy compatibility but it carries higher throughput costs and lacks several Persistent_2 efficiency improvements. Migrate Persistent_1 to Persistent_2 at the next maintenance window.