How to select a storage specification for a General-purpose CPFS file system - Cloud Parallel File Storage

General-purpose Cloud Parallel File Storage (CPFS) supports two storage specifications: a 100 MB/s/TiB baseline and a 200 MB/s/TiB baseline. This topic describes the performance metrics, applicable advanced features, and data liquidity performance for file systems that have the same capacity but different specifications. This information helps you select a suitable storage specification for your General-purpose CPFS file system as needed.

File system instance performance

You can access General-purpose CPFS file systems using CPFS-POSIX or CPFS-NFS clients. The following sections describe the throughput, input/output operations per second (IOPS), and I/O latency of General-purpose CPFS file systems with 100 MB/s/TiB and 200 MB/s/TiB baselines for the same file system capacity.

CPFS-POSIX

After you create a CPFS POSIX mount target, you can access the file system using a CPFS-POSIX client.

Important

The performance of your file system varies based on its creation date, version number, and region.
For more information about how to view the version number of a file system, see View the file system version number.

Metric	100 MB/s/TiB baseline	200 MB/s/TiB baseline
File system version number	v2.3.4 and later	v2.3.4 and later
Supported regions	China (Hangzhou) China (Shanghai) China (Beijing)	China (Hangzhou) China (Shanghai) China (Beijing)
Throughput	min{0.1 × storage capacity (GiB), 30,000} MB/s	min{0.2 × storage capacity (GiB), 45,000} MB/s Up to 100,000 MB/s. To request a higher throughput capacity, submit a ticket.
IOPS	min{15 × storage capacity (GiB), 3,600,000}	min{30 × storage capacity (GiB), 7,200,000} Up to 10,000,000. To request a higher IOPS, submit a ticket.
Average single-stream 4k read latency	0.6 ms	0.4 ms
Average single-stream 4k write latency	0.8 ms	0.6 ms

CPFS-NFS

After you enable the CPFS protocol service, you can use a CPFS-NFS client to access a General-purpose CPFS file system. The CPFS protocol service provides NFS protocol mount targets that are independent of POSIX mount targets. General-purpose CPFS provides two types of NFSv3 protocol services: General-purpose and Cache-accelerated. The following table compares the metrics of the two protocol service types. You can select a suitable protocol service type to access your file system as needed.

Protocol service type	Metric	100 MB/s/TiB baseline specification	200 MB/s/TiB baseline specification
	Throughput	Same as the CPFS file system bandwidth, min[100 × storage capacity (TiB), 20,000] MB/s.	Same as the CPFS file system bandwidth, min[200 × storage capacity (TiB), 20,000] MB/s.
	IOPS	Read/Write: min[4,000 × storage capacity (TiB), 960,000]	Read/Write: min[8,000 × storage capacity (TiB), 960,000]
	Single-stream latency	Read latency: 0.8 ms Write latency: 0.8 ms	Read latency: 0.6 ms Write latency: 0.6 ms
	Maximum throughput per client	600 MB/s	600 MB/s
	Maximum IOPS per client	Read: 80,000 IOPS Write: 30,000 IOPS	Read: 80,000 IOPS Write: 30,000 IOPS
Cache-accelerated	Throughput	The maximum read bandwidth specified when you create the protocol service. The maximum value is 100 GB/s.	The maximum read bandwidth specified when you create the protocol service. The maximum value is 100 GB/s.
	IOPS	Read: min[7,500 × protocol service bandwidth (GB/s), 15,000 × storage capacity (TiB), 1,900,000] Write: min[4,000 × storage capacity (TiB), 960,000]	Read: min[7,500 × protocol service bandwidth (GB/s), 30,000 × storage capacity (TiB), 1,900,000] Write: min[8,000 × storage capacity (TiB), 960,000]
	Single-stream latency	Read latency (cache hit): 0.2 ms Read latency (cache miss): 0.8 ms Write latency: 0.8 ms	Read latency (cache hit): 0.2 ms Read latency (cache miss): 0.6 ms Write latency: 0.6 ms
	Maximum throughput per client	1.1 GB/s	1.1 GB/s
	Maximum IOPS per client	Read: 90,000 IOPS Write: 40,000 IOPS	Read: 90,000 IOPS Write: 40,000 IOPS

Recommendations for choosing between CPFS-POSIX and CPFS-NFS

If you do not have extreme performance requirements, we recommend that you enable the General-purpose protocol service in the console and use a CPFS-NFS client to access your General-purpose CPFS file system for a better user experience.

CPFS-NFS supports the following operating systems:

Operating system	Version
Alibaba Cloud Linux	Alibaba Cloud Linux 3.2104 64-bit Alibaba Cloud Linux 2.1903 64-bit
CentOS	CentOS 8.x 64-bit CentOS 7.x 64-bit
Ubuntu	Ubuntu 20.04 64-bit Ubuntu 18.04 64-bit Ubuntu 16.04 64-bit
Debian	Debian 10.x 64-bit Debian 9.x 64-bit Debian 8.x 64-bit

If you have extreme performance requirements and meet all the following conditions, you can use a CPFS-POSIX client:

You do not rely on the Container Storage Interface (CSI) to access CPFS.
You do not use mmap.

The operating system is one of the following versions.

Operating system	Distribution	Kernel version
Alibaba Cloud Linux	Alibaba Cloud Linux 2.1903 64-bit	4.19.91-27.4.al7.x86_64 and earlier versions
Red Hat Enterprise Linux (RHEL) or CentOS	8.4	4.18.0-305.19.1.el8_4
	8.3	4.18.0-240.22.1.el8_3
	8.2	4.18.0-193.28.1.el8_2
	8.1	4.18.0-147.8.1.el8_1
	8.0	4.18.0-80.11.2.el8_0
	7.9	3.10.0-1160.42.2.el7
	7.8	3.10.0-1127.19.1.el7
	7.7	3.10.0-1062.18.1.el7
	7.6	3.10.0-957.54.1.el7
	7.5	3.10.0-862.14.4.el7
	7.4	3.10.0-693.2.2.el7
	7.3	3.10.0-514.26.2.el7
	7.2	3.10.0-514.26.2.el7
Ubuntu	20.04.3 LTS	5.4.0-86-generic

Data liquidity performance

General-purpose CPFS supports data liquidity with Object Storage Service (OSS). When you create a data liquidity task, the General-purpose CPFS file system automatically synchronizes object metadata from the specified OSS bucket. After the synchronization is complete, you can quickly process data from OSS using a high-performance, POSIX-compatible file interface. General-purpose CPFS file systems also allow you to export data to an OSS bucket using the CPFS console or OpenAPI. The following table describes the performance.

Operation	Metric	Description
Data import	The throughput of files that are larger than 1 GB.	The throughput of a single file is 200 MB/s. The throughput of multiple files reaches the level of bandwidth.
Data import	The IOPS of files that are larger than 1 MB.	The IOPS of one or more directories is 1000.
Data export	The throughput of files that are larger than 1 GB.	The throughput of a single file is 200 MB/s. The throughput of multiple files reaches the level of bandwidth.
Data export	The IOPS of files that are larger than 1 MB.	The IOPS of one or more directories is 600.
Data deletion	OPS	The IOPS of one or more directories is 2000.
On-demand data loading	The throughput of files that are larger than 1 GB.	The throughput of a single file is 200 MB/s. The throughput of multiple files reaches the level of bandwidth.
On-demand data loading	The IOPS of files that are larger than 1 MB.	The IOPS of one or more directories is 1000.
Automatic metadata update	OPS	Dataflows at 600 MB/s: 2000. Dataflows at 1200 MB/s: 3000. Dataflows at 1500 MB/s: 4000.

Capacity specifications

Starting capacity: 3,600 GiB
Scaling step size: 1,200 GiB
Maximum capacity: 1 PiB

Cloud Parallel File Storage:Product specifications