tencent cloud

Feedback

Storage Types and Performance

Last updated: 2022-03-16 15:26:44

    CFS provides file systems with scalable storage, which can work with other Tencent Cloud services such as CVM, TKE, and BatchCompute. CFS offers four storage types, including Standard, Standard Turbo, High-Performance, and High-Performance Turbo. You can select one to fit your business needs.

    Standard

    Standard is a highly cost-effective file system that uses mixed media and accelerates data reads/writes through a data tiering mechanism. Three replicas on three independent physical servers on different racks are provided to guarantee strong consistency and successful storage of every data entry written to the file system. Its access server features hot data migration to ensure data reliability and service high availability, making it suitable for scenarios that require small-scale general data storage.

    Standard Turbo

    Standard Turbo is a parallel file system that uses mixed media and an asymmetric framework. The data nodes and metadata nodes are deployed independently. By allowing mounting with a private protocol, the performance of a single client can be like a storage cluster. In addition, underlying resources are isolated to ensure exclusive storage for the cluster. Three replicas on three independent physical servers on different racks are provided to guarantee strong consistency and successful storage of every data entry written to the file system. Its access server features hot data migration to ensure data reliability and service high availability, making it suitable for scenarios that require large-scale throughput and mixed loads.

    High-Performance

    High-Performance is a low-latency file system that uses NVMe only and provides a high storage performance through a data tiering mechanism. Three replicas on three independent physical servers on different racks are provided to guarantee strong consistency and successful storage of every data entry written to the file system. Its access server features hot data migration to ensure data reliability and service high availability, making it suitable for small-scale core businesses that are latency-sensitive.

    High-Performance Turbo

    High-Performance Turbo is a high-bandwidth, low-latency, parallel file system that uses NVMe only and an asymmetric framework. The data nodes and metadata nodes are deployed independently. By allowing mounting with a private protocol, the performance of a single client can be like a storage cluster. In addition, underlying resources are isolated to ensure exclusive storage for the cluster. Three replicas on three independent physical servers on different racks are provided to guarantee strong consistency and successful storage of every data entry written to the file system. Its access server features hot data migration to ensure data reliability and service high availability, making it suitable for scenarios that use large numbers of small files.

    Product Type Standard High-Performance
    Product Name Standard Standard Turbo High-Performance High-Performance Turbo
    Positioning Cost-effective, suitable for small-scale general data storage High-throughput and large storage, suitable for businesses that require high throughput and mixed loads High-performance and low latency, suitable for small-scale core businesses that are latency-sensitive High-throughput and high IOPS, suitable for businesses that use large-scale small files
    Scenario Small-scale enterprise file sharing, data backup/archive, and log storage Non-linear media asset editing, image rendering, AI inferencing, OLAP business, and high-performance computing Small-scale CI/CD development and testing environments, high-performance web services, OLTP databases, and high-performance file sharing High-performance and large-scale computation, AI training, OLTP databases, big data analysis, and OLAP services
    Storage capacity 0−160 TiB 40 TiB−100 PiB 0−32 TiB 20 TiB−100 PiB
    Bandwidth (MiB/s) Min {100 + 0.1 × Capacity in GiB , 300} Min{0.1 × Capacity in GiB , 100,000} Min{200 + 0.2 × Capacity in GiB , 1024} Min{0.2 × Capacity in GiB , 100,000}
    Read IOPS min{2000 + 8 × Size in GiB , 15000} Min{2 × Capacity in GiB , 2 million} Min{2500 + 30 × Capacity in GiB , 30,000} Min{20 × Capacity in GiB , 10 million}
    Write IOPS min{2000 + 8 × Size in GiB , 15,000} Min{1 × Capacity in GiB , 1 million} Min{2500 + 30 × Capacity in GiB , 30,000} Min{5 × Capacity in GiB , 3 million}
    OPS upper threshold Read/Write: 10,000/1,000 Read/Write: 300,000/20,000 Read/Write: 30,000/3,000 Read/Write: 300,000/20,000
    Latency 4K single-stream read:3 ms
    4K single-stream write: 7 ms
    4K single-stream read: 0.2 ms
    4K single-stream write: 3 ms
    4K single-stream read: 1 ms
    4K single-stream write: 1.5 ms
    4K single-stream read: 0.2 ms
    4K single-stream write: 1.5 ms
    Cost 0.058 USD/GiB/month 0.09 USD/GiB/month 0.23 USD/GiB/month 0.15 USD/GiB/month
    Supported protocol NFS/SMB POSIX/MPI NFS POSIX/MPI
    Expansion Auto Manual Auto Manual
    Supported OS Linux/Windows Linux Linux/Windows Linux

    Description

    • In the performance-related formulas, the capacities of Standard Turbo and High-Performance Turbo refer to the capacities purchased for the cluster. For Standard and High-Performance, the capacities refer to the storage that is actually used by the instances.
    • The table above shows the capabilities of the file system. To reach the performance upper threshold, you usually need to perform multi-threaded reads/writes using multiple compute nodes.
    • The performance benchmark is tested in interruption-free conditions. The results of mixed tests or other loads might vary.
    • OPS suggests the file system's ability to process metadata per second, which is not the same as IOPS.
    Contact Us

    Contact our sales team or business advisors to help your business.

    Technical Support

    Open a ticket if you're looking for further assistance. Our Ticket is 7x24 avaliable.

    7x24 Phone Support