DEV Community

Cover image for 2026-11-22: I deployed third-generation FSx for ONTAP in us-east-1. Deployable Regions and documented limits that disagree

2026-11-22: I deployed third-generation FSx for ONTAP in us-east-1. Deployable Regions and documented limits that disagree

Figure 1: Multi-AZ 3 and Single-AZ 3 in us-east-1, reached by an Amazon EC2 client over NFS, SMB and iSCSI. The right column lists the per-HA-pair limits, the Regions where it could be deployed, and unit prices

Key points (checked 2026-11-22):

  • I deployed third-generation file systems (SINGLE_AZ_3, MULTI_AZ_3) in us-east-1
  • Of seven Regions I checked, only three US Regions were judged deployable. That disagrees with the user guide's Region table, which also lists Tokyo, Spain and Ireland
  • For workloads where many servers read and write at once, or that need high IOPS (chip design and verification, rendering, databases on Multi-AZ), more fits in one file system
  • For the same deployment form, generation 2 and 3 have the same unit prices. What changes is the ceiling: with 12 HA pairs, Single-AZ 3 allows 221,184 MBps of read throughput and 8,640,000 SSD IOPS
  • You can check whether a Region can deploy it without creating anything (steps below)

In November 2026, Amazon FSx for NetApp ONTAP began supporting third-generation file systems.

The documentation lists up to 18,432 MBps and 720,000 SSD IOPS per HA pair. Reading it did not settle two questions:

  • Which Regions can deploy the new deployment types
  • How the price changes from the second generation

So I deployed both types in us-east-1 and checked.

Who gets more options with generation 3

The What's New announcement names chip design and verification, media rendering and video editing, and mission-critical databases as fits for generation 3. These are workloads where many servers read and write the same data at once, or where one database needs high IOPS. The higher ceilings change the picture as follows (values are documented; I did not measure performance).

Who Constraint until now What generation 3 changes
Teams running chip design and verification (EDA) on AWS Many compute nodes reading the same libraries and projects approach the generation 2 read ceiling of 73,728 MBps at 12 HA pairs One file system can go up to 221,184 MBps of reads. Extending from on-premises ONTAP to the cloud with FlexCache is also an AWS reference architecture
Render farms and video editing When render nodes read assets all at once, the bandwidth ceiling can become wait time The per-pair ceiling rises from 6,144 to 18,432 MBps, so fewer HA pairs give the same bandwidth
Teams running databases on Multi-AZ Multi-AZ is one HA pair, so generation 2 caps it at 200,000 SSD IOPS and 6,144 MBps Multi-AZ 3 goes up to 720,000 SSD IOPS and 18,432 MBps, so cross-AZ high availability and high IOPS fit in one file system more often
Teams that added HA pairs for bandwidth 73,728 MBps needs 12 HA pairs on generation 2 The same 73,728 MBps fits in 4 HA pairs. The SSD minimum is 1,024 GiB per HA pair, so the minimum capacity falls from 12 TiB to 4 TiB (quotas page)

As a database example, AWS documents SAP HANA on FSx for ONTAP. I did not check whether generation 3 is covered. Check each product's requirements and supported configurations before adopting it.

What this post covers, and what it does not

Covers:

  • Which Regions could deploy it, and how that differs from the user guide's Region table
  • What I saw when creating Single-AZ 3 and Multi-AZ 3 (creation time, automatic IOPS, response fields)
  • A stale IOPS limit in the API reference
  • Monthly price comparison across generations in us-east-1 for the same capacity

Does not cover:

  • Real throughput or IOPS. I did not benchmark; the limits are from the user guide
  • Failover behavior and scale-up time
  • Regions other than the seven I checked (us-east-1, us-east-2, us-west-2, ap-northeast-1, eu-west-1, eu-south-2, ap-southeast-1)
  • A basis for procurement or contracts. Which Regions can deploy it is one day of results; repeat the check right before you rely on it

The previous post (FSx for ONTAP and S3 Files, measured, Japanese) measured performance. This one checks availability and where the sources disagree.

Environment

Item Value
Checked on 2026-11-22 (JST). Later changes are not checked
Regions us-east-1 for deployment and pricing; seven Regions checked for deployability
Created SINGLE_AZ_3 and MULTI_AZ_3, one HA pair each, 1,024 GiB SSD, 384 MBps
Interface AWS CLI and signed SDK requests (the public SDK model already contains MULTI_AZ_3 and SINGLE_AZ_3)
Cleanup File systems, VPC and secrets deleted after testing; the KMS key is pending deletion

Fewer Regions could deploy it than the user guide lists

I sent SINGLE_AZ_3 to all seven Regions and MULTI_AZ_3 to Tokyo only. The seven Regions are not exhaustive, and I did not record a selection criterion.

I judged deployability by sending CreateFileSystem with a subnet that does not exist. No file system is created.

  • If the Region can deploy generation 3: the deployment type passes validation and the call stops at "subnet does not exist"
  • If it cannot: the call fails with BadRequest
Source (read on 2026-11-22) Third-generation Regions
AWS What's New announcement US East (N. Virginia, Ohio), US West (Oregon)
User guide Region table (Availability by AWS Region) The three above, plus Asia Pacific (Tokyo), Europe (Spain), Europe (Ireland)
Result of the check Table below
Region Verdict Actually deployed?
us-east-1 Deployable Yes (two file systems). All deployment results below come from this Region
us-east-2, us-west-2 Deployable No (check only)
ap-northeast-1, eu-west-1, eu-south-2, ap-southeast-1 Not deployable (BadRequest) —

The response does not say why BadRequest came back. Only Tokyo had a control: the same request with SINGLE_AZ_2 passed there, and MULTI_AZ_3 was rejected. For the other three Regions I did not rule out a cause other than the deployment type, such as per-account allow-listing. I also cannot tell whether the table was updated early or the launch is still rolling out. If a design depends on a Region, send the request there and see.

Third-generation deployment results

Item Result
Time to AVAILABLE About 13 min (MULTI_AZ_3) and 13.5 min (SINGLE_AZ_3), one run each
Minimum throughput 384 MBps for one HA pair
SSD IOPS Automatic: 3,072 for 1,024 GiB (3 per GiB; one capacity only)
Response fields ThroughputCapacityPerHAPair and HAPairs

The console create screen lists "Multi-AZ 3 (recommended)" and "Single-AZ 3" as deployment types.

File system create screen listing the Multi-AZ 3 and Single-AZ 3 deployment types (English console, us-east-1)

The limits, per HA pair and per file system

The table shows configurable maximums, not measured performance. It comes from the performance page and the quotas page (read 2026-11-22). The per-file-system values are for 12 HA pairs on Single-AZ and one HA pair on Multi-AZ.

Deployment type HA pairs Read throughput (per pair / file system) Write throughput (per pair / file system) SSD IOPS (per pair / file system) SSD capacity (file system)
SINGLE_AZ_3 1 to 12 18,432 / 221,184 MBps 4,500 / 54,000 MBps 720,000 / 8,640,000 1 PiB
MULTI_AZ_3 1 18,432 / 18,432 MBps 4,500 / 4,500 MBps 720,000 / 720,000 512 TiB
SINGLE_AZ_2 (for comparison) 1 to 12 6,144 / 73,728 MBps 2,048 / 24,576 MBps 200,000 / 2,400,000 1 PiB
MULTI_AZ_2 (for comparison) 1 6,144 / 6,144 MBps 2,048 / 2,048 MBps 200,000 / 200,000 512 TiB
  • The file-system SSD IOPS is the documented "720,000 per HA pair (up to 12 pairs)" multiplied by 12
  • What's New's "216 GB/s" is 221,184 MBps divided by 1,024, and "18 GB/s" is 18,432 MBps. Convert to MBps before comparing

For HPC, EDA and rendering, where many clients read at once, size from the file-system column. With 12 HA pairs, Single-AZ 3 raises the read ceiling to three times generation 2 (73,728 to 221,184 MBps) and SSD IOPS to 3.6 times (2,400,000 to 8,640,000).

Check these conditions before planning near those ceilings:

  • Account quotas: by default each account and Region allows 10,240 MBps of total throughput capacity and 1,000,000 SSD IOPS. Even one HA pair at its maximum (18,432 MBps) exceeds the default, so request an increase
  • HA pairs and throughput: with two or more HA pairs, choose at least 1,536 MBps per pair. 384 and 768 MBps are for one pair only
  • Writes: the write ceiling is about a quarter of the read ceiling. Size write-heavy work from the write columns
  • Block storage: iSCSI and NVMe/TCP are available with six or fewer HA pairs
  • Clients: a single network flow is limited to 5 Gbps. Use NFS nconnect or SMB Multichannel to add connections (per the performance page)

I verified only the IOPS limit. Sending UpdateFileSystem with Iops 720,001 to a one-pair third-generation file system returned the message below. The API reference for DiskIopsConfiguration says the maximum is 200,000 × HAPairs, which has not caught up with generation 3.

Invalid SSD IOPS provided: 720001 IOPS. The maximum SSD IOPS per ONTAP file system is 720000.
Enter fullscreen mode Exit fullscreen mode

I sent only an over-limit value, so nothing was billed. I did not check the limits for multiple HA pairs.

Second and third generation cost the same per unit

Unit prices come from the Pricing API (us-east-1, fetched 2026-11-22), for 1,024 GiB SSD, automatic IOPS, one HA pair and 730 hours. Backups, capacity pool storage and data transfer are not included.

Configuration Throughput Monthly
Gen 1 Single-AZ 128 MBps (minimum) about $220
Gen 1 Single-AZ 512 MBps about $497
Gen 2 / Gen 3 Single-AZ 384 MBps (minimum) about $742
Gen 1 Multi-AZ 128 MBps (minimum) about $410
Gen 1 Multi-AZ 512 MBps about $870
Gen 2 / Gen 3 Multi-AZ 384 MBps (minimum) about $1,216

The rows have different throughput, so do not read the price gaps as a generation difference. Compare unit prices:

Deployment form SSD (per GB-month) Throughput (per MBps-month)
Gen 1 Single-AZ $0.125 $0.72
Gen 1 Multi-AZ $0.25 $1.20
Gen 2 / Gen 3 Single-AZ $0.125 $1.60
Gen 2 / Gen 3 Multi-AZ $0.25 $2.50

For the same deployment form, generations 2 and 3 have the same unit prices. What changes is the ceiling (18,432 MBps per pair against 6,144 MBps); I did not measure any performance difference. Generation 1 has a lower throughput price and a different set of selectable values. The jump in minimum throughput from 128 to 384 MBps is what shows up in small configurations.

Check it in your own environment

  1. In the Region you want, send CreateFileSystem with a non-existent subnet for SINGLE_AZ_3 and MULTI_AZ_3. No file system is created, and the response tells you whether the Region can deploy the type. Use a subnet ID that clearly does not exist:
aws fsx create-file-system --region <region> --file-system-type ONTAP \
  --storage-capacity 1024 --subnet-ids subnet-0123456789abcdef0 \
  --ontap-configuration DeploymentType=SINGLE_AZ_3,ThroughputCapacityPerHAPair=384,HAPairs=1
Enter fullscreen mode Exit fullscreen mode

InvalidNetworkSettings (with InvalidSubnetId in the details) means the Region can deploy it; BadRequest means it cannot. Use the same request with SINGLE_AZ_2 as a control

  1. Pull the ONTAP unit prices with aws pricing get-products for AmazonFSx
  2. After creation, read ThroughputCapacityPerHAPair and HAPairs from describe-file-systems

Details and conditions are in a note in my GitHub repository (FSx for ONTAP Adoption Playbook; the note is in Japanese):

What I did not measure

  • Throughput and IOPS with data on the volumes; the file systems were deleted after the checks
  • Whether Regions beyond the seven checked can deploy it
  • When the Region table will match the Regions that can deploy it

I hope this saves some checking for anyone evaluating third-generation file systems.

Bye now!!

Yoshiki Fujiwara

Top comments (0)