Technical Presentation: Fabricpool in Ontap 9.12.1
Technical Presentation: Fabricpool in Ontap 9.12.1
John Lantz
Senior Technical Marketing Engineer
March 2023
FabricPool
Combines performance and cloud tier
into one storage pool that manages data seamlessly
and reduces the total cost of ownership
3 © 2022 NetApp, Inc. All rights reserved. — NETAPP CONFIDENTIAL —
The FabricPool Solution
What Is FabricPool?
Automatic tiering
Zero-touch management
On-Premises Footprint
Preserves file system
FabricPool
Lower cost of ownership
Local Tier
Cloud Tier
80% Inactive
Data
Object Storage
Before After
Active Data Inactive Data
4 © 2022 NetApp, Inc. All rights reserved. — NETAPP CONFIDENTIAL —
End-to-End Security
What Is FabricPool?
Compression
Deduplication
Compaction
www.evaluatorgroup.com/FabricPool/
Capacity
Growth
Snapshots
Cold data
Multi-year TCO
Local
Hybrid Cloud
Tier Smart Economics
Transparent to existing
On Premises TCO savings applications
Cloud
Tier
Simple Data Security
Public Cloud Private Cloud
Quick setup On premises
Little policy Over the wire
management Cloud
Before
• All flash or hybrid with solid-state drive (SSD)
Primary Cluster aggregates
Provisioned storage
Active data
Snapshot copies
Inactive data
• snapshot show
Provisioned storage
Active data
Snapshot copies S3
Inactive data
Before After • Auto volume tiering policy tiers all inactive (cold)
Primary Cluster Primary Cluster blocks
Provisioned storage
• -tiering-minimum-cooling-days
Active data
Snapshot™ copies S3
Inactive data
Provisioned storage
Active data
Inactive data
Expand capacity
NetApp®
SnapMirror®
No change to data
management
NetApp
SnapVault®
Provisioned storage
Active data
S3
Inactive data
NetApp®
SnapMirror® Limits
Tiering stops when aggregate
reaches 98% capacity or tiers 16PB
NetApp
SnapVault® 392TB local + 7.8PB cloud
400TB aggr
Provisioned storage @5% metadata usage
Active data
S3
Inactive data
• Requirements
• Configuration
• Best practices
• Key takeaways
First available in ONTAP 9.10.1, SVM Migrate allows cluster administrators can non-disruptively relocate an
SVM from a source cluster to a destination cluster to manage capacity and load balancing, or to enable
equipment upgrades or data center consolidations.
Note: SVM Migrate is only supported on AFF arrays. Although support for FabricPool was added in 9.12.1,
there is still a lengthy list of features that do not support SVM Migrate.
NetApp
SnapMirror Beginning in 9.12.1, customers can
Disaster use all three features together,
Recovery®
protecting SVMs hosting FlexGroup
Provisioned storage
Active data
volumes on FabricPool aggregates.
S3
Inactive data
Attach cloud tier storage (Amazon S3, Microsoft Azure Blob Storage, StorageGRID®, etc.) to aggregates
• Snapshot-Only
Cold NetApp® Snapshot blocks not shared by the active file system are moved to the cloud tier
• Auto
Cold blocks are moved to the cloud tier
• All
Blocks are tiered directly to the cloud tier; no temperature scans are required
• None (default)
Cold blocks are not moved to the cloud tier
23 © 2022 NetApp, Inc. All rights reserved. — NETAPP CONFIDENTIAL —
Block Temperature
How does FabricPool work?
NetApp® FlexVol® Volume 3 Tiering scan collects cold blocks ( ) and packages them into 4MB
objects (1,024 4K blocks)
NetApp® FlexVol® Volume 3 Tiering scan collects cold blocks ( ) and packages them into 4MB
objects (1,024 4K blocks)
Note: Tiering does not take place until the aggregate is at more
than 50% capacity
• Cold blocks are fetched using a GET operation, written to the performance tier, and made hot … but not in all
circumstances:
• Snapshot-Only
Read: cold blocks are made hot and written to the performance tier
• Auto
Random read: cold blocks are made hot and written to the performance tier
Sequential read: cold blocks stay cold; blocks are not written to the performance tier
• All
Read: cold blocks stay cold; blocks are not written to the performance tier
• Supported platforms
• Thin-provisioned volume
• Certificate authority (CA) certificate (StorageGRID®, other private clouds, including ONTAP S3)
• The amount of space grows as data is added instead of being determined by the initial volume size
• This growth is essential for FabricPool because the volume must support external cloud tier data that
becomes hot and is brought back to the local tier
• Quality of service (QoS) throughput floors, QoS Min, guarantees that throughput for a workload does not fall
below a minimum number of IOPS
• Solution
• Turn off QoS Min on volumes you want to tier
• Turn off tiering (-tiering-policy none) on volumes that need QoS Min
• Private clouds, such as NetApp® StorageGRID®, or ONTAP S3, require security certificates for securing
access to:
• The management interface
• Storage and API gateway nodes
• DNS hostname to Data LIF resolution
• CA certificates should be installed on NetApp ONTAP® before a private object store bucket is attached to
an aggregate
• Failing to do so will result in an error unless validation has been turned off
(not recommended)
-is-certificate-validation-enabled false
• Configure Amazon S3
• Create bucket, purchase license, and install license
• Configure ONTAP S3
• Please refer to:
S3 Configuration Power Guide: http://docs.netapp.com/ontap-9/topic/com.netapp.doc.pow-s3-cg/S3%20configuration.pdf
• Attach aggregate
• Create a bucket
1. Click Create bucket.
2. Enter a unique DNS-compliant bucket name.
3. Select a region close to your data center.
4. Click Create.
• Additional settings
• Do not edit bucket properties or permissions.
• When ordering a new cluster, sales should include a free 10TB FabricPool license;
for existing systems, a FabricPool license must be purchased
Note: A license is not required when using NetApp® StorageGRID® as the cloud tier
or Amazon S3 as the cloud tier in Cloud Volumes ONTAP.
• FabricPool licenses are clusterwide; customers need their cluster UUID when obtaining a license (cluster
identity show)
• For additional info about point-of-sale and add-on sales, refer to the ONTAP FabricPool Licensing Overview
and Understanding the FabricPool License
• FabricPool licenses are available in perpetual and term-based (1- or 3-year) formats
• After obtaining a NetApp® license file (NLF), install it using NetApp OnCommand® System Manager
• Create a container
1. Click Settings > Containers.
2. Enter a name.
3. Select an access level.
4. Click OK.
• When ordering a new cluster, sales should include a free 10TB FabricPool license;
for existing systems, a FabricPool license must be purchased
• A license is not required when using NetApp® StorageGRID® as the cloud tier
or Amazon S3 as the cloud tier in ONTAP® Cloud
• FabricPool licenses are clusterwide; customers need their cluster UUID when obtaining a license (cluster
identity show)
• For additional info about point-of-sale and add-on sales, refer to the ONTAP FabricPool Licensing Overview
• FabricPool licenses are available in perpetual and term-based (1- or 3-year) formats
• After obtaining a NetApp® license file (NLF), install it using NetApp OnCommand® System Manager
• Connect to StorageGRID
1. Click File.
2. Select New S3 Compatible Account.
3. Select S3 Compatible Account.
4. Enter the display name, service point, access key, and
secret key associated with your StorageGRID
account.
5. Click Test Connection.
6. Click OK.
Locate CA certificate
• StorageGRID® admin
1. Select Configuration > Grid Options.
• Identification and attachment of a cloud tier can be completed by using either NetApp® OnCommand®
System Manager or NetApp ONTAP® CLI
2. Click Storage.
6. View and update the tiering policies for the volumes on the aggregate.
Note: By default, volume tiering policies are set as Snapshot-Only.
7. Click Save.
• Snapshot-Only
• Cold blocks associated with NetApp® Snapshot copies that are not shared with the active file system will be moved to the cloud tier
• Auto
• Cold blocks are moved to the cloud tier
• The default setting for tiering-minimum-cooling-days is 31 days; adjustable from 2 to 183 days (63 days max prior to 9.8)
• All
• All blocks are immediately moved to the cloud tier (WAFL® metadata stays on the local tier.)
• Warning!
• Tiering data immediately reduces storage efficiencies. (Inline only, no background efficiencies. Use Auto 2-days to gain background effeciencies)
• Object storage is not transactional and will result in significant fragmentation if subjected to change.
• None (default)
• Cold blocks in the volume are not moved to the cloud tier
• Previously tiered data remains on the cloud tier until it is read and made hot
2. Select a volume.
5. Click Save.
volume modify
-vserver <name>
-volume <name>
-tiering-policy <snapshot-only, auto, all, none (default)>
-tiering-minimum-cooling-days <#>
• Performance
• Connectivity
• Buckets
• Data redundancy
• Storage classes
• Data protection
• NetApp® StorageGRID®
• More nodes = more CPU
• More CPU = more IOPS
• More IOPS = more GETs
• Public Clouds
• IOPS limited by provider
• Number of concurrent sessions
• Constrained at the bucket level
? 1000*
65 © 2022 NetApp, Inc. All rights reserved. — NETAPP CONFIDENTIAL —
Performance (Continued)
Best Practices
• Reading data from the cloud tier is not as fast as reading data from SSDs on the local tier (MBps vs. GBps)
Note: Adjust recovery time objective (RTO) accordingly.
• Although FabricPool and your chosen object store can withstand 10-second latency, client applications might
not be able to
• Do not run virtualized StorageGRID, or other virtualized object stores, in volumes that tier inactive data. Set
tiering policy to None.
• Cold blocks on the cloud tier become unavailable until it comes back online
SAN UNRECOVERED READ ERROR/ The application might need to be restarted so that the read can
RECOMMEND REWRITE THE DATA be retried – please read the slide notes
• Bucket creation
• Amazon S3
• Microsoft Azure Blob Storage
Aggr_1
• NetApp® StorageGRID®
• Information lifecycle management (ILM): tiering and deletion polices on the object
store
• Other than StorageGRID® data replication and erasure coding, and changing
storage classes in public clouds, FabricPool does not support ILM policies applied to
object store buckets
Note: Starting in ONTAP 9.8, using tags, FabricPool supports ILM policies on
StorageGRID.
• NetApp® StorageGRID®
• 2+1 erasure coding
• 99.9999999999999% durability (15 nines)
• Amazon S3
• Automatically replicates to multiple locations
• 99.999999999% durability (11 nines)
• ONTAP S3
• No additional data replication other than standard ONTAP data protection
• Recommend using SnapMirror or other data protection technologies
• Amazon S3
• Standard
• Standard - Infrequent Access
• One Zone
• Intelligent-Tiering
• Commercial Cloud Services (C2S)
• Traditional aggregate
Aggregate
Vol1
Vol2
Vol3
• FabricPool aggregate
Aggregate
+ Vol1
Cloud
Vol2
Vol3
• FabricPool aggregate
Aggregate
+ Vol1
Cloud
Vol2
Vol3
Aggregate
+ Vol1
Cloud NetApp®
SnapMirror®
Vol2
Vol3
Source Destination
75 © 2022 NetApp, Inc. All rights reserved. — NETAPP CONFIDENTIAL —
Capacity
Best practices
• Sizing
• Hot data
• Cold data’s metadata
• Cold data that is made hot 1:10
• Cold data until it is tiered to the cloud tier
• Limits
• Tiering stops when an aggregate reaches 98% capacity
or when 16PB has been tiered from the aggregate,
whichever comes first %
• 800TB 784TB at 98%
• @ 5% metadata, 784TB SSD will tier 15.6PB
98 16PB
• 400TB 392TB at 98%
• @ 5% metadata, 392TB HDD will tier 7.8PB
• ONTAP Local
50TB
+ Cloud
980TB
50TB
+ Cloud
318TB
• 50 + 318 = 368
77 © 2022 NetApp, Inc. All rights reserved. — NETAPP CONFIDENTIAL —
Supported Features
NetApp ONTAP 9.8
Encryption
NetApp Volume Encryption (NVE)
NetApp Storage Encryption (NSE)
• FabricPool collection:
https://fieldportal.netapp.com/collections/533113
The Amazon S3 Glacier Instant Retrieval storage class uses a different cost structure but provides the same
performance as the S3 Standard storage class.
Note: Glacier Flexible Retrieval and Glacier Deep Archive are not supported.
Put throttling allows storage administrators to set an upper threshold on the maximum, per node, PUT rate.
PUT throttling can be useful when network resources, or the object store endpoint, are resource constrained.
Although rare, resource constraints can occur with underpowered object stores or during the first days of
FabricPool usage when TB or PB of cold data begin to tier out.
Note: FabricPool PUTs do not compete for resources with other applications. FabricPool PUTs are
automatically placed at a lower priority (bullied) by client apps and other ONTAP workloads, such as
SnapMirror. PUT throttling using put-rate-limit is unrelated to concurrent traffic.
Temperature sensitive storage efficiency (TSSE) was introduced in ONTAP 9.8. With this feature, storage
efficiency is applied according to whether data is hot or cold. For cold data, 32KB data blocks are
compressed, and for hot data, which is overwritten more often, smaller data blocks are compressed, making
the process more efficient.
Compression is a trade-off of space saved versus the performance impact to compress and de-compress the
data, TSSE automatically applies the optimal level of space saving dependent on how hot the data is.
Due to various issues, prior to 9.10.1, volumes with TSSE would tier uncompressed data, losing storage
efficiencies associated with compression on objects in the cloud tier. For this reason, TSSE was not enabled
by default on volumes in FabricPool aggregates.
ONTAP 9.10.1 supports a unified object format in which volumes with TSSE will preserve compression-based
storage efficiencies when tiering … and provide the additional storage efficiencies associated with TSSE.
• ONTAP S3
• Cloud Retrieval
• Object Tagging
• HDD Support
• 9.2 – 9.7 SSD only
• 9.6 added HDD support for inactive data reporting
• Primary environments
• Does tiering from HDD to HDD make sense?
• Migrations become much easier with
optimized volume moves
• Cold data tiers before the migration
• Only hot data needs to be provisioned
Vol
• Secondary environments
• Prior to 9.8, SSD aggregates were required even in
backup/DR environments
(WAFL metadata always stays on the local tier)
• With the release of 9.8, the local tier can be provisioned with
HDD aggregates
• StorageGRID
• Preferred choice for FabricPool*
• More data tiered to StorageGRID than any other S3 target
• Robust distributed \ multisite capable S3 object storage
• Positions NetApp to support additional S3 workloads
• volume modify
-vserver <vserver-name> -volume <volume-name>
-cloud-retrieval-policy
default = random reads only
on-read = random and sequential reads
never = cold blocks stay cold
promote = brings data back on the next tiering scan
None: 31-days
Snapshot-only: -tiering-minimum-cooling-days
Auto: -tiering-minimum-cooling-days
All: All user data is inactive
StorageGRID 40% NA
• MetroCluster Support
• NDMP Support
• S3 Compatible Providers
• Aggregate-to-bucket relationships
• 1:1
• Many:1
Aggr_1
• 1:Many (not OK)
Aggr_1
Aggr_2
Aggr_1
• Aggregate-to-bucket relationships
• 1:1
• Many:1
Aggr_1
• 1:Many
Aggr_1
• Multi-vendor solutions
• Mirror the same data to multiple vendors for an
additional level of resiliency.
Aggr_1
Cluster A Cluster B
Site A Site B
• Continuous availability.
• Redundant fabrics.
Cluster A Cluster B
Site A Site B
• Continuous availability.
• Redundant fabrics.
A1 A2
B2 B1
ONTAP 9.7
• NDMP operations now supported on FabricPool aggregates
Note: Customers who use third party backup tools should configure them to use NDMP as the protocol—not SMB or NFS.
ONTAP will treat backups operations using NAS protocols as client reads—making the data hot. With frequent backups, it is likely that the
data being backed up will never become cool enough to tier.
NDMP reads are treated differently (ONTAP uses FPolicy hooks and does not consider them as client reads) and will not keep the data
hot.
ONTAP 9.7
• FPVR process no longer required.
• Customer will still need to perform the testing and confirm that the
object store meets their requirements.
• If it does, FabricPool will be supported.
• SVM-DR
• Works like the Backup volume tiering policy, but can be used on non-DP volumes.
• Blocks in the volume are immediately marked as cold ( ) and tiered.
• NetApp WAFL® metadata is always hot. ( )
• If read, cold blocks stay cold and are not written back to the performance tier.
• Warning!
• Tiering data immediately reduces storage efficiencies. (Inline only)
• Only use on volumes you are confident contain cold data that will not change.
• Object storage is not transactional and will result in significant fragmentation if subjected to change.
• Volumes with active RW workloads should use the Auto (with an appropriate tiering-minimum-cooling-days setting) volume tiering policy
instead of the All Volume tiering policy.
Volume moves now only need to move the WAFL metadata on the
performance tier.
Tiered data does not need to move, reducing network traffic and
associated costs.
ONTAP 9.2-9.5
Cold data is brought
back to the cluster …
Aggr_1
Vol1 Bucket 1
… then tiered back
to the cloud
Aggr_2
• ONTAP 9.6
Aggr_1
Cold data
Vol1 never leaves Bucket 1
the bucket!
Aggr_2
Assuming no other data lives on an 800TB aggregate and all volumes use the All tiering policy:
Dataset using 5% metadata = x20 tiered = 15.6PB tiered + 784TB on SSD
Dataset using 10% metadata = x10 tiered = 7.8PB tiered + 784TB on SSD
Dataset using 20% metadata = x5 tiered = 3.9PB tiered + 784TB on SSD
Advantages include:
OpEx, not CapEx
No additional support cost line items
Costs amortized over time rather than large immediate expense
• FlexGroup Volumes
• Client-Side Encryption
+ Cloud
• Requirements
• All aggregates used by the FlexGroup volume must be FabricPool aggregates
• FlexGroup volumes use None as the default volume tiering policy
(FlexVol volumes use Snapshot-Only)
• All constituent/member volumes in a FlexGroup volume use the same volume tiering policy
• Standard FabricPool requirements apply (SSDs, thin provisioning, etc.)
1. Set the tiering policy on the existing FlexGroup to the desired value
Additional resources
• Automatic
• No additional configuration needed
• Requirements
• CAP url
• Server CA certificate
• Client CA certificate
• Client private key
• Previous to ONTAP 9.5, no tiering would take place in an aggregate until the performance tier of
the aggregate was at least 50% full.
• In ONTAP 9.5, the tiering fullness threshold is adjustable
• Consider reducing the threshold to reduce the amount of data stored on the performance tier.
• Consider increasing the threshold to tier only when necessary to reduced used capacity of the performance tier.
• Object defragmentation
• Compaction
• CLI
volume modify
-vserver <name>
-volume <name>
-tiering-policy <snapshot-only (default), auto, backup, none>
-tiering-minimum-cooling-days <#>
• ONTAP anywhere
• ONTAP
• Cloud Volumes ONTAP
• ONTAP Select
S3
• ONTAP Select
• NetApp recommends using SSDs for FabricPool aggregates
• Premium license supports SSD aggregates
• Non-FabricPool aggregates
• Snapshot™ only 31 days
• Auto 31 days
• Backup 31 days
• None 31 days
• FabricPool aggregates
• Snapshot only 31 days
• Auto 31 days (default; tiering-minimum-cooling-days)
• Backup Not reported
• None 31 days
Test latency and throughput performance of object stores before you attach them to FabricPool aggregates
Start
storage aggregate
object-store
profiler start Op Size Total Failed Latency(ms) Throughput
-object-store-name <name> min max avg
-node <name> -------------------------------------------------------------------------------
PUT 4MB 50000 0 273 9500 719 277.3MB
• FabricPool now defrags objects based on the percentage of used blocks in the object
• Defrag reclaims space used by blocks that have been deleted or are no longer being referenced by the performance tier
• Prior to NetApp® ONTAP® 9.4, object store cleanup effectively took place at 0% referenced
• Consider reducing defrag thresholds when using Infrequent Access/Cool pricing schemes
• CLI
• storage aggregate object-store modify –aggregate <name> -object-store-name <name> –unreclaimed-space-
threshold <%> (0%-99%)
• Optional on FAS
• Private cloud appliances require unique security certificates for securing access to:
• The management interface
• Storage and API gateway nodes
• StorageGRID® certificate authority (CA) certificates should be installed on ONTAP® before a StorageGRID
bucket is attached to an aggregate
• Failing to do so will result in an error
-is-certificate-validation-enabled false
ONTAP 9.6
Depreciated and replaced by
All Volume Tiering Policy
148 © 2022 NetApp, Inc. All rights reserved. — NETAPP CONFIDENTIAL —
The Backup Policy Tiers Immediately
NetApp® SnapMirror®
vol1
vol1
NetApp SnapVault®
NetApp® SnapMirror®
2 Replicated blocks are immediately made cold ( )
and moved to the cloud tier
vol1
vol1
NetApp WAFL® metadata is always hot ( )
NetApp SnapVault®
NetApp® SnapMirror®
2 Replicated blocks are immediately made cold ( )
and moved to the external cloud tier
vol1
vol1
NetApp WAFL® metadata is always hot ( )