Skip to main content
Keystone
3.0

Performance service levels in Keystone

Contributors netapp-manini netapp-shwetav netapp-aruldeepa netapp-bcammett
PDFs

Keystone STaaS offers pre-defined performance service levels (PSLs) for each storage type. Each PSL specifies the performance targets (IOPS, throughput, and latency), supported protocols, platform, and minimum capacity requirements. Review the available PSLs to select the right service level for your workload and plan your subscription.

A Keystone subscription can include multiple PSLs across one or more storage types. Each PSL has a committed capacity, which is the minimum capacity billed for that service level during the subscription term.

Note An AFX subscription is limited to the AFX storage type and currently offers only one PSL, Extreme.

Performance service levels instances

A performance service level (PSL) can have multiple instances, called performance service level instances (PSLIs), in your environment. Each PSLI is a separate storage system assigned to that performance service level, with committed capacity measured per PSLI.

  • For unified or block-optimized storage, a PSLI is a storage array consisting of two storage controllers in an HA pair with internal or externally attached storage media.

  • For AFX storage, a PSLI is a single AFX storage controller. AFX uses a disaggregated architecture where controllers (compute) and drive shelves (storage) are independent and can be scaled separately. IOPS and throughput targets apply per controller; committed capacity is measured at the AFX cluster level, which includes all controllers and drive shelves in the cluster.

Performance metrics such as IOPS, throughput (GBps), and latency (ms) are measured per PSLI.

Tip A detailed service description for NetApp Keystone STaaS performance service levels is available here.

The base performance service levels for the supported data storage types: unified, block-optimized, AFX, object, and cloud services are described in the following sections:

Performance service levels for unified storage

Performance service level
(all specifications per performance service level instance [HA pair])

Extreme

Premium

Standard

Value

Supported protocols

FC, iSCSI, NFS, NFSv4/RDMA, NVMe/FC, NVMe/TCP, SMB, S3

Sample workload types

AI/ML, HPC, InMem DB

Analytics, EDA, OLTP

OLAP, IoT, Containers

Backup, Archive

Target 90th percentile latency

<=1 ms

<=1 ms

<=4 ms

>4 ms

Target maximum IOPS1

1M

550K

250K

NA

Target maximum GBps1

40

20

10

NA

Platform

AFF A-Series

AFF A-Series

AFF C-Series

FAS

Minimum committed capacity

50 TiB

50 TiB

100 TiB

100 TiB

Incremental committed capacity increase

25 TiB

Committed and metered capacity type

Logical or physical

Performance service levels for block-optimized storage

Performance service level
(all specifications per performance service level instance [HA pair])

Extreme

Premium

Standard

Supported protocols

NVMe/TCP, NVMe/FC, FC, iSCSI

Sample workload types

SAP HANA, Oracle, MS SQL Server, EPIC

Target 90th percentile latency

<=1 ms

<=1 ms

<=4 ms

Target maximum IOPS1

850K

450K

120K

Target maximum GBps

65

25

6

Platform

ASA A-Series

ASA A-Series

ASA C-Series

Minimum committed capacity

50 TiB

50 TiB

100 TiB

Incremental committed capacity increase

25TiB

Committed and metered capacity type

Logical or physical

Performance service levels for AFX storage

Performance service level
(all specifications per performance service level instance [one AFX storage controller])

Extreme

Supported protocols

NFSv3, NFSv4/RDMA, pNFS, SMB, S3

Sample workload types

AI/ML, HPC, InMem DB

Target 90th percentile latency

<=1 ms

Target maximum IOPS1

625K

Target maximum GBps

20

Platform

AFX

Minimum committed capacity per AFX cluster2

200 TiB

Incremental committed capacity increase

100 TiB

Committed and metered capacity type

Physical
Note

  • 1Mutually exclusive targets. Actual performance may differ based on various factors, including the operating system version, hardware, workload type, and number of concurrent operations.

  • 2In an AFX cluster, performance targets apply per storage controller. Each controller added to the cluster contributes the same IOPS and throughput. Committed capacity and incremental increases apply to the total drive shelf capacity across the cluster.

Performance measurement conditions for unified, block-optimized, and AFX storage

The base performance service level metrics depend on the following conditions:

  • The performance service levels support ONTAP 9.8 and later.

  • For unified storage,

    • IOPS: For ONTAP 9.16.1 with NFS, each performance level instance supports random access with a 70% read and 30% write ratio, an 8 KB block size, and a latency of 1 ms (4 ms for Standard).

    • Throughput: For ONTAP 9.16.1 with NFS, each performance level instance supports sequential access with 100% read and a 32 KB block size.

  • For block-optimized storage,

    • IOPS: For ONTAP 9.16.1 with FCP, each performance level instance supports random access with a 70% read and 30% write ratio, an 8 KB block size, and a latency of 1 ms.

    • Throughput: For ONTAP 9.16.1 with FCP, each performance level instance supports sequential access with 100% read and a 64 KB block size.

  • For AFX storage,

    • IOPS: For ONTAP 9.18.1 with NFSv4, each performance level instance supports random access with a 70% read and 30% write ratio, and a 4 KB block size.

    • Throughput: For ONTAP 9.18.1 with NFSv4, each performance level instance supports sequential access with 100% read and a 64 KB block size.

  • Latency does not include the following:

    • application or host latency

    • customer network latency to or from the controller ports

    • overheads associated with the data transfer to the object store in case of FabricPool

  • Latency values are not applicable to MetroCluster write operations. These write operations are dependent on the distance of remote systems.

  • Expected IOPS is targeted for FabricPool only if the tiering policy is set to "none" and no blocks are in the cloud. Expected IOPS is targeted for volumes that are not in a SnapMirror synchronous relationship.

Performance service levels for object storage

Performance service level

Standard

Value

Supported protocol

S3

Platform

SGF6112

SG6160

Minimum committed capacity per order

200 TiB

500 TiB

Incremental committed capacity increase

25 TiB

100 TiB

Committed and metered capacity type

Physical

Cloud storage

Performance service level

Cloud Volumes ONTAP

Supported protocols

NFS, CIFS, iSCSI, and S3 (AWS and Azure only)

Platform

Cloud Volumes ONTAP

Minimum committed capacity per order

4 TiB

Incremental committed capacity increase

1 TiB

Committed and metered capacity type

Logical
Note

  • Cloud native services, such as compute, storage, networking, are invoiced by cloud providers.

  • These services are dependent on cloud storage and compute characteristics.

Related information