TL,DR (at the end too): Azure NetApp Files should be your default File Share solution in Azure, it is lower cost and high performing that Azure Files Premium.
Six months ago, Cool Access was introduced to Azure NetApp Files (ANF), bringing a huge cost-saving impact for ANF customers. Yet, many Azure engineers have yet to take notice. Historically, here’s how Azure Storage Services have been perceived:
Pre-Cool Access Market Perspectives on Azure Storage Services:
| Options | Capacity Cost | Transactions Cost | Market Perspective | |
| Azure Files Standard (HDD) | Transaction Optimized/Hot/Cool | $15-$61/TiB/Mo | $0.015-$0.13/10k | “Slow and cheap” |
| Azure Files Premium (SSD) | Premium | $164/TiB/Mo | Included | “Fast enough and not too expensive” |
| Azure NetApp Files (SSD) | Standard/Premium/Ultra | $151-$402/TiB/Mo | Included | “Fastest but you pay a premium“ |
How ANF Cool Data Access Works
Cool Access automatically moves inactive data to cheaper blob storage while keeping it seamlessly accessible to users and applications. When accessed, ANF transparently retrieves the data from cool storage, ensuring a smooth user experience.
The biggest difference? Cost savings. Instead of paying a fixed per-TiB price for an ANF capacity pool, data stored in Cool Access is billed at a much lower rate (e.g., $0.06 vs. $0.39/GiB/Month). Active data remains on ANF’s primary storage, charged at the regular service level rates.
For complete pricing details, check out: Azure NetApp Files Pricing.
How Cool Access Has Changed the Market
Most file share workloads contain at least 75% cold data, meaning Azure customers can realize substantial cost savings by moving file shares, VDI profiles, and other workloads to ANF with Cool Access.
This change completely reshuffles the storage pricing landscape:
Post-Cool Access: Market Perspective on Azure Storage Services
| Storage Tier | Options | Capacity Cost | Transactions Cost | Market Perspective |
| Azure Files Standard (HDD) | Transaction Optimized/Hot/Cool | $15-$61/TiB/Mo | $0.015-$0.13/10k | “Slow and cheap” |
| Azure Files Premium (SSD) | Premium | $164/TiB/Mo | Included | “Fast enough but costs more than ANF“ |
| Azure NetApp Files (SSD) | Standard/Premium/Ultra | $85-$148/TiB/Mo | Included | “Fastest, without the premium cost“ |
Comparing Costs: A 10 TiB Deployment
Below is a cost comparison for 10 TiB of storage in East US, before considering snapshot savings:
| Storage Tier | Capacity | Hot Data | Cool Data | Unused Data | Snapshot | Total Cost ($/Mo) |
| ANF Standard | 10 TiBs | $377 | $475 | $0 | $0 | $853 |
| ANF Premium | 10 TiBs | $753 | $475 | $0 | $0 | $1,228 |
| ANF Ultra | 10 TiBs | $1,005 | $475 | $0 | $0 | $1,480 |
| AF Premium | 10 TiBs | $1,638 | $0 | $0 | $0 | $1,638 |
Even without leveraging ANF’s snapshot technology, ANF is already more cost-effective than Azure Files Premium.
Why a column for “Unused Data” and “Snapshots”?
When we do more realistic modeling, we’ll often need to provision more capacity to provide increased throughput. In these situations, there is unused capacity that also incurs a cost, in my modeling spreadsheets I include this. Additionally, the NetApp snapshot technology included with ANF is a significant advantage for the ANF solution.
The Advantage of ANF Snapshots
ANF’s snapshot technology is a key differentiator. Unlike Azure Files snapshots, which are file-based, ANF minimizes storage usage by leveraging a block-level, redirect-based snapshot system. This allows for:
- Efficient data replication
- Low-cost snapshot storage on Cool Access
- Minimal impact on overall storage consumption
Cost Comparison: Snapshots & Backups
| Backup Type | Azure NetApp Files Cost | Azure Files Premium Cost |
|---|---|---|
| Snapshots | $62–$146/TiB/Month (cool tier) | $139/TiB/Month |
| Backups | $51.20/TiB/Month | $139/TiB/Month |
ANF Snapshot Technology (and Backup Costs)
ANF’s snapshot technology is unique, by using a redirect index for each snapshot and targeting data blocks and not files, the amount of storage used for snapshots in minimized. The awareness of data block locations on a disk also enables the efficient data replication and cool data tiering features. Details on this technology can be found here: How Azure NetApp Files snapshots work | Microsoft Learn
Combining cool access data tiering and ANF’s snapshot technology the majority of snapshot data ends up on the cool tier. Assuming a 28-day snapshot retention and a 7-day coolness period, the blended cost for ANF snaps is $84-$146/TiB/Mo and only applies to the very efficient deltas. A longer retention period and shorted coolness period can lower this to as low as $62/TiB/Mo.
Alternatively, Azure Files Premium snapshots are $139/TiB/Mo (East US) without any cool data tiering.
Backups are even more of a difference. ANF backups cost $51.20/TiB/Mo while Azure Files Premium uses the same costs as their snapshot above: $139/TiB/Mo
A More Realistic Cost Estimate with Snapshots
Let’s assume our 10TiB deployment also needs 4 weeks of snapshots and 1 year of monthly backups. We’ll also need to assume a daily data change rate of 1% so we can calculate backup data consumption.
| Storage Tier | Capacity | Hot Data | Cool Data | Unused Data | Snapshot | Total Cost (Mo) |
| ANF Standard | 13 TiBs | $377 | $475 | $36 | $231 | $1,120 |
| ANF Premium | 13 TiBs | $753 | $475 | $72 | $335 | $1,635 |
| ANF Ultra | 13 TiBs | $1,005 | $475 | $96 | $405 | $1,981 |
| AF Premium | 10 TiBs +13 TiBs Snaps | $1,638 | $0 | $0 | $1,777 | $3,416 |
An Apples-to-Apples Comparison by Matching Throughput Performance
The other consideration is available throughput for the workload. Each of these four options have their own unique throughput allocation based on the capacity allocated to the service. In this example, the AF Premium option offers the most MiB/s as configured above. While the use-case does not necessarily need that throughput (and the final architecture should reflect that requirement), in order to get an apples-to-apples comparison, the following is the cost structure when the throughput of each option is the same or higher than Azure Files Premium.
But not quite “Apples-to-Apples” Since ANF IOPS are Uncapped
Azure Files Premium has an IOPS limit based on allocated capacity. For the examples below, Azure Files Premium is limited to 13,240 IOPS (but can burst up to 30,720 IOPS for short durations. ANF on the other hand does not cap IOPS. In the examples below every ANF service level supports 290,000+ (4k random) IOPS. 9.44x higher sustained IOPS than the “burstable” limit of Azure Files, nearly 22x more IOPS vs. Azure Files sustained IOPS.
Also, latency is different and no amount of capacity or increased throughput will change that. In my experience, Azure NetApp Files is typically delivering ~3ms (E2E) latency while Azure Files Premium is in the 10-13ms range.
| Storage Tier | Capacity | Throughput | Hot Data | Cool Data | Unused Data | Snapshot | Total Cost (Mo) |
| ANF Standard | 71 TiBs | 1,136 MiB/s | $377 | $475 | $8,794 | $231 | $9,878 |
| ANF Premium | 32 TiBs | 1,152 MiB/s | $753 | $475 | $5,796 | $335 | $7,359 |
| ANF Ultra | 16 TiBs | 1,152 MiB/s | $1,005 | $475 | $1,302 | $405 | $3,187 |
| AF Premium | 10 TiBs +13 TiBs Snaps | 1,124 MiB/s | $1,638 | $0 | $0 | $1,777 | $3,416 |
How about using that unused data for cool data storage?
Cool access creates a situation where higher data consumption causes lower costs. Unused space in your ANF Capacity pool is billed as “hot” data while used data that is cool is charged at a lower rate. Thus, filling your volume with data, even if you don’t “need” it on that volume, can lower your costs.
Based on how the cool access data is billed additional savings can be found by using the “unused” space with data that will go cool. Unused data is billed at the hot cost, while used & cool data is billed at the lower rate. Options include moving organizational archive data, generating filler data, and/or extending snapshot retention. Thus, assuming the excess data capacity can be used and tiered to cool, the costs drop further:
| Storage Tier | Capacity | Throughput | Hot Data | Cool Data | Unused Data* | Snapshot | Total Cost (Mo) |
| ANF Standard | 71 TiBs | 1,136 MiB/s | $377 | $475 | $3,570 | $231 | $4,654 |
| ANF Premium | 32 TiBs | 1,152 MiB/s | $753 | $475 | $1,179 | $335 | $2,742 |
| ANF Ultra | 16 TiBs | 1,152 MiB/s | $1,005 | $475 | $198 | $405 | $2,084 |
| AF Premium | 10 TiBs +13 TiBs Snaps | 1,124 MiB/s | $1,638 | $0 | $0 | $1,777 | $3,416 |
* Unused Data refers to the space that is occupied by data stored in ANF for the purpose of lowering costs.
Let’s keep going and use workload shaping to save even more money
ANF performance can be increased and decreased non-disruptively. Let’s extend our example scenario and assume it is an Azure Virtual Desktop file share or profiles storage service. Let’s also assume the majority of the users work a standard work week (Active hours: Monday to Friday, 6am-6pm).
What can we do to further reduce costs? There are several options, not all can be used together, depending on your situation.
- Weekend Service Level Shaping: Reduce the Service Level on weekends (6pm Friday -> 6am Monday). That is 60 hours per week, representing 35.7% of the week.
- Daily Pool Capacity Shaping: Reduce the size of the Capacity Pool on evenings and weekends (6pm->6am even night + all weekend). That is 108 hours per week, representing 64.3% of the week.
- Coolness Period Weekend Fast Forward: Adjust the coolness period Friday nights so that cool data is tiered down sooner. This is a very minimal savings strategy but fun as a mental exercise.
Weekend Service Level Shaping
An example PowerShell script for this can be found on the Awesome ANF resources site.
You can move volumes between service levels (Standard, Premium & Ultra) non-disruptively and via automation, including PowerShell as illustrated above. The only limitation is that a volume must stay in a higher level for 24 hours before dropping back down.
By dropping from Ultra Standard over the weekend, the throughput would drop to 256MiB/s but that is still 22% of the peak throughput during a weekend when we expect <5% of the workload.
The ANF Ultra cost would be reduced because the standard rate would be charged on those 16 TiBs for 35.7% of the time. The result would be a reduction from $2,084 to $1,798/Mo
| Storage Tier | Capacity | Throughput | Hot Data | Cool Data | Unused Data | Snapshot | Total Cost (Mo) |
| ANF Ultra Active Hours: 108 | 16 TiBs * 108hrs | 1,152 MiB/s | $647 | $305 | $128 | $260 | $1,340 |
| ANF Standard Inactive Hours: 60 | 16 TiBs * 60hrs | 256 MiB/s | $135 | $170 | $71 | $83 | $458 |
| Total: $1,798 |
Daily Pool Capacity Shaping
Another strategy is to reduce the capacity (and thus throughput) of the deployment on evenings and weekends.
In the above example the ANF Ultra tier has 16 TiB deployed, while only 13 TiB is needed for the data size. That capacity pool can be reduced to 13TiB on evenings and weekends. For this approach you’d not be able to keep the excess data in a cool tier, so you’d need to start with the example where ANF Ultra is $3,187 monthly.
By reducing the size of the volume for 64.3% of the time you’d end up with a cost structure like this:
| Storage Tier | Capacity | Throughput | Hot Data | Cool Data | Unused Data | Snapshot | Total Cost (Mo) |
| ANF Ultra Active Hours: 60 | 16 TiBs * 60hrs | 1,152 MiB/s | $359 | $170 | $71 | $145 | $746 |
| ANF Ultra Inactive Hours: 108 | 13 TiBs *108hrs | 720 MiB/s | $646 | $305 | $10 | $260 | $1,221 |
| Total: $1,967 |
Coolness Period Weekend Fast Forward
Admittedly this is getting a bit ridiculous, at some point there is a diminishing return on the complexity and detail needed to implement these things… I think this next approach is far over that line. But I’m a nerd and enjoy spreadsheets so here it goes…
Starting with the assumption that weekend activity is close to zero, it follows that if a block of hot data is within 2 days of going to cool on Friday night, it is nearly certain that it will move to cool lover the weekend.
We can expedite the move of that data from hot to cool by automating a change in coolness period each Friday evening and Monday morning.
Each Friday at 6pm, we’ll automate the reduction in the coolness period by 2 days. This will cause any hot data that will move to cool over the weekend immediately move rather than waiting up to 2 extra days to move. On average this will move data to the cool tier 30 hours earlier that it would otherwise (since the weekend is 60 hours and data hotness is evenly distributed).
Then on Monday morning at 6am we’ll return the coolness period to the original value.
Moving data from the ANF Ultra tier to the cool data tier 30 hours early will save $14/TiB of data moved. In our example we have 10 TiBs of data, with a 1% daily data change rate. Of that, 75% of the changed data is assumed to move to cool. Meaning that 2.25 TiB’s of data is always on the pipeline from hot to cool. 64.3% of this data will move to cool naturally on a weekday, but the remaining 35.7% of the data (.8TiBs) would be tiered over the weekend.
So, by accelerating that data to cool every Friday night, we’ll save approximately $11.25 each weekend, or $48.73/month. It works out to 1.5% savings vs. Cost Chart 3 – Capacity, Snapshots and Throughput.
Yes, this seems like a lot of complexity to save a few dollars, but for some large deployments this might be worth the effort. For example, this strategy could save $4,990/month on a 1PiB deployment, enough to pay a Jr. technician to manage the complexity, even as a full-time job (which it’s not).
Combine Weekend, Daily and Coolness Fast Forward?
Yes, it is possible. No, I’m not going to do the math. No, you probably won’t do it in the real world.
These last couple examples are based on so many assumptions that they’re only useful to illustrate the relative impact (order of magnitude?) of each strategy. With each additional strategy layered on this, the complexity increases and the impact decreases. There is overlap in the savings achieved by each. For example, fast forwarding the cool data assumes you drop From Ultra to Cool (a delta of $0.33/GiB/Mo) but if you have the weekend shaping strategy in effect, the savings would be from Standard to Cool (a delta of $0.088/GiB/Mo).
Too Long, Didn’t Read!
Yea, I hear you, I have a hard time re-reading it myself. A wall of text is a big bummer.
Bottom line: Azure NetApp Files should be your default File Share solution in Azure:
- ANF is not a 3rd party/marketplace service. It is built, sold and supported by Microsoft.
- ANF is the “Enterprise” Storage Service for Azure
- ANF delivers latency 3x lower than Azure Files Premium >10ms (Azure Files) vs.<3ms (ANF). ANF can even be configured to deliver sub-millisecond latency!
- SAP HANA, Microsoft SQL, Oracle databases shine on ANF because of the low latency.
- ANF costs less than Azure Files Premium for most real-world use-cases:
- Need raw storage? ANF Standard is $151/TiB vs. Azure Files Premium at $164/TiB.
- Got a standard file share? 4 TiBs of ANF Premium with 60% cool data is $634/Mo vs. 4TiBs of Azure Files Premium at $655/mo
- Storing FSLogix profiles? Cool data tiering won’t slow down logins and data re-heats faster than users can notice. ANF outperforms Azure Files Premium and typically costs 1/3rd after snapshots, backups and cool data savings all stack up.
- Need to take snapshots? (yes, you do!) ANF snaps on cool storage can be <$0.06/GiB vs. $0.136/GiB for Azure Files Premium snapshots.
- Backups? (you better be backing up that data!) ANF Backups are $0.05/GiB vs. $0.136/GiB for Azure Files Premium snapshots.
- Need extreme IOPS? ANF offers uncapped IOPS, as much IO as you can fit in the available throughput. With Azure Files Premium, both the Throughput and IOPS are capped based on capacity. ANF Standard offers more IOPS that Azure Files Premium at all capacity levels above 1 TiB.