Updated on 2025-05-07 GMT+08:00

Storage Service Selection

Huawei Cloud provides Object Storage Service (OBS), Elastic Volume Service (EVS), and Scalable File Service (SFS). The following table compares the three types of storage services.

Table 1 Comparison between the three types of storage services

Dimension

EVS

SFS

OBS

Description

High reliability, high IOPS, and elastic scalability (equivalent to disks)

High bandwidth, on-demand expansion, and shared access (equivalent to NAS)

High reliability, low cost, massive scalability, and support for objects of any type and size

Scenario

HPC, enterprise core cluster applications, enterprise application systems, and development and testing

High-performance computing (HPC), media processing, file sharing, content management, and web services

Big data analytics, static website hosting, online video on demand (VoD), gene sequencing, and intelligent video surveillance

Storage logic

Stores binary data and cannot directly store files. To store files, you need to format the disk with a file system first.

SFS stores files, and sorts and displays in the hierarchy of files and folders.

Stores objects. Files are saved directly to OBS. The files automatically generate corresponding system metadata. You can also customize the metadata if needed.

Access method

It can only be used and accessed from applications after being attached to ECSs or BMSs and initialized (OS layer, application reconstruction not involved).

It can be mounted to ECSs, BMSs, or CCEs using network protocols. NFS and CIFS are supported (CIFS is not supported by common file systems). A network address must be specified or mapped to a local directory for access. (OS layer, application reconstruction is not involved).

OBS buckets can be accessed through the Internet or Direct Connect. The bucket address must be specified for access using HTTP or HTTPS. (Application layer, the application needs to integrate the SDK or invoke APIs, which involves application reconstruction.)

Data sharing

Supported. It is controlled by the cluster management software installed on the ECS/BMS and cannot be shared across AZs.

Supported. You can directly access SFS using NFSv3 (SFS Turbo also supports CIFS). Cross-AZ sharing is supported.

Supported. Direct HTTP/HTTPS access is required, and unlimited sharing is supported.

Remote access

Not supported

Supported

Supported

Independent use

Not supported

Supported

Supported

Capacity

TiB-level

PiB-level (SFS Trubo)/EiB-level (general purpose file system)

EiB-level

Latency

Minimum

Medium

Maximum

Bandwidth (throughput)

MiB/s-level

GiB/s-level

TiB/s-level

Data redundancy

Single AZ

Single AZ (SFS Turbo)/Single or multiple AZs (general purpose file system)

Single AZ/Multiple AZs

Data reliability

9 nines

10 nines

11 nines in a single AZ or 12 nines in multiple AZs

Storage billing mode

Pay-for-capacity

SFS Turbo billed by capacity/SFS general file system billed by usage

Pay-as-you-go

The following describes the selection principles of storage services.

Service Applicability Principles

You need to select a proper storage type based on service scenarios. The following aspects must be considered:

  1. Available access mode: After EVS disks or SFS file systems are attached to a host, they are displayed as file system paths in the OS and can be directly accessed by upper-layer applications. OBS needs to be accessed by service applications using dedicated SDKs or APIs. You need to know the acceptable access modes of services. For database applications that require direct raw disk mapping, only block storage (EVS) can be used.
  2. Shared or not: EVS supports sharing. You need to select the sharing feature during purchase and use the dedicated cluster software to manage shared disks. SFS and OBS support sharing. Therefore, you need to analyze whether the content to be stored needs to be shared by multiple nodes based on service scenarios.
  3. Storage capacity: Different storage types support different capacities. You need to estimate the required capacity level based on the current service volume and future development to select a proper storage type.
    Table 2 Minimum and maximum capacity of the storage service

    Storage Type

    Minimum Capacity

    Maximum Capacity

    EVS

    10 GB

    32 TB

    SFS Turbo

    20 MB/s/TiB

    3.6 TB

    1 PB

    Others

    1.2 TB

    1 PB

    SFS general-purpose capacity-oriented

    0

    Unlimited

    OBS

    0

    Unlimited

Performance Matching

The performance metrics of storage services include transmission bandwidth, IOPS, and latency, as shown in the following table. You need to select the appropriate storage service and specifications based on the performance requirements and specifications of the service system.

In addition, EVS and OBS have no restrictions on the size of stored objects. SFS general-purpose Capacity-Oriented is not suitable for applications with massive small files smaller than 1 MB. SFS Turbo and subsequent SFS General Performance-Oriented can support massive small file applications.

Table 3 Performance metrics of storage services

Storage Type

Bandwidth Upper Limit (GB/s)

IOPS Upper Limit

Average Latency

EVS

High I/O-SAS

0.15

5K

1–3 ms

General-purpose SSD-GPSSD

0.25

20K

1 ms

Ultra-high I/O-SSD

0.35

50K

1 ms

General Purpose SSD V2-GPSSD2

1

128K

1 ms

Extreme SSD-ESSD

1

128K

Sub-milliseconds

Extreme SSD V2-ESSD2

4

256K

Sub-milliseconds

SFS Turbo

20 MB/s/TiB

20

250K

2–5 ms

40 MB/s/TiB

20

250K

2–5 ms

125MB/s/TiB

100

Millions

1–3 ms

250 MB/s/TiB

100

Millions

1–3 ms

500 MB/s/TiB

200

Millions

1–3 ms

1,000 MB/s/TiB

200

Millions

1–3 ms

SFS General-Purpose

Capacity

50

100K

7 ms

Performance

200

2000K

5 ms

OBS

TB-level

Tens of millions

10 ms+

Cost Optimization

You need to consider costs when selecting a storage type to reduce storage costs while meeting service performance requirements.

  1. Select the storage service with the lowest unit price while meeting service performance requirements.
  2. For storage (EVS and SFS Turbo) billed by specifications, predict the service increment and monitor the capacity. You are advised to reserve 15% to 20% as the scale-out threshold to prevent resource waste caused by excessive capacity specifications.
  3. For pay-per-use storage (general-purpose SFS and OBS), plan the usage and purchase resource packages to reduce costs.
  4. You can plan lifecycle management policies for storage (general-purpose SFS and OBS) and move cold data to infrequently accessed storage in a timely manner to reduce costs.
  5. For services that require large storage capacity and long data retention period, you can reconstruct the service application layer and combine different types of storage (for example, combine EVS/SFS Turbo and OBS) to optimize costs while ensuring service performance.

Reliability Assurance

EVS, SFS Turbo, general-purpose SFS, and OBS use three-replica storage. Data durability meets service requirements, but reliability varies.

  • The three replicas of EVS and SFS Turbo are in the same AZ. If the AZ egress or equipment room is faulty, services will be unavailable.
  • General-purpose SFS and OBS support both single-AZ and multi-AZ deployment. (Currently, general-purpose SFS supports only single-AZ deployment. Multi-AZ deployment will be available in the future.) For services that require high continuity, you can select multi-AZ instances.
  • EVS supports quick data backup and restoration using images, snapshots, and cloud backup. SFS Turbo supports backup and restoration using cloud backup. General-purpose SFS and OBS are generally used in ultra-large-capacity service scenarios and the backup capability is not planned.

Based on the preceding selection principles, the following are some typical scenario suggestions:

  • Shared disks are not recommended unless self-built databases are deployed in two-node cluster or cluster mode. Instead, SFS is used to implement file sharing among multiple hosts. (Shared disks cannot be attached to multiple ECSs across AZs, but SFS supports this function.)
  • For applications that require frequent read and write of a large number of logs and log summary analysis, SFS is recommended as the unified log storage for multiple nodes (select the type based on performance requirements).
  • For asynchronous interaction and latency-insensitive services, OBS is preferred to reduce costs. If services are difficult to adapt to reconstruction, SFS general-purpose capacity-oriented storage can be used.
  • In AI scenarios, it is recommended that SFS Turbo and OBS be used together to reduce costs and improve performance.