CPFS product introduction-Cloud Parallel File Storage(CPFS)-阿里云帮助中心

This topic introduces the editions of Cloud Parallel File Storage (CPFS), compares them, and explains how to get started with CPFS.

What is CPFS

Cloud Parallel File Storage (CPFS) is a fully managed and scalable parallel file system from Alibaba Cloud, designed for high-performance computing (HPC) workloads. CPFS provides a unified namespace that allows hundreds or thousands of machines to access data concurrently. It delivers tens of GB/s in throughput and millions of IOPS while maintaining sub-millisecond latency.

CPFS editions

CPFS offers the following editions:

CPFS General-purpose Edition: Ideal for HPC workloads that demand high throughput, high IOPS, and support for massive files. Common use cases include AI training, autonomous driving, genomics, media rendering, EDA simulation, oil and gas exploration, and meteorological analysis.
CPFS for Lingjun: Optimized for intelligent computing scenarios such as AIGC and autonomous driving.
Currently, CPFS for Lingjun is compatible only with PAI-Lingjun intelligent computing resources, PAI general computing resources, PAI-Lingjun intelligent computing resources (single-tenant), and Container Compute Service (ACS). It cannot be accessed from ECS instances.

How to choose: If you use PAI or ACS and need extreme performance, choose CPFS for Lingjun. If you need to mount the file system directly on an ECS instance or use the NFS v3 protocol, choose CPFS General-purpose Edition.

Edition comparison

CPFS for Lingjun and CPFS General-purpose Edition have different target customers, use cases, features, and limits, as detailed in the following table.

Category	Comparison criteria	CPFS for Lingjun (invitational preview)	CPFS General-purpose Edition
Positioning	Target customers	Intelligent computing workloads	HPC workloads
	Product features	Ultra-high throughput and IOPS with an end-to-end RDMA network	High throughput and IOPS
	Purchase method	Requires whitelisting for purchase.	Available for direct purchase.
Purchase	Available specifications	400 MB/s/TiB baseline	100 MB/s/TiB baseline 200 MB/s/TiB baseline
Usage	Protocol	POSIX	Interoperability between POSIX and NFS v3 protocols
	Client	Linux	Linux
	Region	Public cloud: China (Hangzhou) China (Shanghai) China (Shenzhen) China (Heyuan) China (Guangzhou) (Decommissioned) China (Beijing) China (Ulanqab) China (Hong Kong) Singapore Germany (Frankfurt) Thailand (Bangkok) Malaysia (Kuala Lumpur) Dubai Japan (Tokyo) Malaysia (Johor) Finance cloud: China East 2 Finance	China (Hangzhou) China (Shanghai) China (Qingdao) China (Beijing) China (Zhangjiakou) China (Hohhot) China (Ulanqab) China (Shenzhen) China (Heyuan) China (Chengdu) Singapore
	Capacity specifications	Starting capacity: 10 TiB Maximum capacity: 1 PiB To create a file system of up to 20 PiB, submit a ticket.	Starting capacity China (Ulanqab): 4,800 GiB Other regions: 3,600 GiB Maximum capacity: 1 PiB
Performance	Throughput	Read throughput: min(400 × Capacity in TiB, 400,000) MB/s Write throughput: min(200 × Capacity in TiB, 200,000) MB/s The maximum throughput can be increased to 2 TB/s. To request this, submit a ticket.	100 MB/s/TiB baseline: min(0.1 × Capacity in GiB, 30,000) MB/s 200 MB/s/TiB baseline: min(0.2 × Capacity in GiB, 45,000) MB/s The maximum throughput can be increased to 100,000 MB/s. To request this, submit a ticket.
	IOPS	Read IOPS: min(6,800 × Capacity in TiB, 6,800,000) Write IOPS: min(2,300 × Capacity in TiB, 2,300,000) The maximum IOPS can be increased to 30,000,000. To request this, submit a ticket.	100 MB/s/TiB baseline: min(15 × Capacity in GiB, 3,600,000) 200 MB/s/TiB baseline: min(30 × Capacity in GiB, 7,200,000) The maximum IOPS can be increased to 10,000,000. To request this, submit a ticket.
	Average latency for a single 4 KB read	0.25 ms	100 MB/s/TiB baseline: 0.6 ms 200 MB/s/TiB baseline: 0.4 ms
	Average latency for a single 4 KB write	0.6 ms	100 MB/s/TiB baseline: 0.8 ms 200 MB/s/TiB baseline: 0.6 ms
Elasticity	Elastic scale-out	Supported	Supported
	Elastic scale-in	Not supported	Not supported
	Scaling method	Manual	Manual
	Scaling step size	10 TiB	China (Ulanqab): 2,400 GiB Other regions: 1,200 GiB
Security	ACL-based access control	Not supported	Supported
	Server-side encryption	Not supported	Supported
	Encryption in transit	Not supported	Supported
Stability	Subdirectory mounting	Not supported	Supported
	Dataflow	Supported	Supported
	Backup	Supported	Supported
	Locally redundant storage	Supported	Supported
	Zone-redundant storage	Not supported	Supported only for the 100 MB/s/TiB baseline
Cost	Billing method	Pay-as-you-go	Pay-as-you-go resource plan

Get started with CPFS

Task		CPFS for Lingjun (invitational preview)	CPFS General-purpose Edition
Learn about CPFS		What is CPFS for Lingjun (invitational preview)? Specifications Limits Billing	What is CPFS General-purpose Edition Specifications Limits CPFS billing
View specifications		Specifications	Specifications
Create a CPFS file system		Create a file system	Create a file system
Mount a CPFS file system		Mount a CPFS for Lingjun file system in PAI Mount a CPFS for Lingjun file system in ACS	Mount a CPFS file system on an ECS instance Mount a CPFS file system in a container
Use CPFS	Query file system details	Query file system details	Query file system details
	Scale out a file system	Scale out a file system	Scale out a file system
	Protocol service	Not supported	Protocol service
	Dataflow	Dataflow for CPFS for Lingjun (invitational preview)	CPFS Dataflow
	Monitor performance and configure alerts	Data monitoring	Data monitoring
	Migrate data to CPFS	Data migration	Not supported
	Access control	Not supported	Access control
	View audit logs	Log management	Not supported
	Back up CPFS data	Data backup	Data backup
	Zone-redundant storage	Not supported	Storage redundancy
API reference		List of operations by function	API overview