CPFS Parallel File Storage

Benefit

Description

Hybrid Cloud Storage architecture

Integrates with public cloud storage services to provide a high-quality experience for customers in Cloud Bursting scenarios.

High scalability

• A CPFS Parallel File Storage cluster can scale out to a maximum of 16,384 nodes.

• The fully symmetrical distributed architecture allows for linear scaling of throughput for both metadata and data.

• Seamless scale-out: Automatically detects the load on existing storage nodes during scale-out to control the expansion speed.

• Multi-level storage architecture: Supports scaling out on-premises CPFS Parallel File Storage to CPFS and OSS in the cloud.

High performance

• A single CPFS Parallel File Storage cluster supports throughput of up to TB/s.

• Supports 100G Ethernet network access, providing 2.3 GB/s of throughput for a single node.

• Supports 100G/200G InfiniBand network access for system-wide improvements in throughput and latency.

High availability and reliability

• Rolling Update feature: Provides rolling upgrades to ensure uninterrupted service during the upgrade process.

• Second-level fault detection: Quickly detects failed disks and down service nodes.

• Supports multiple data protection modes:

  • Replica mode: 2 replicas, 3 replicas

  • Erasure coding mode: (4+2p), (4+3p), (8+2p), (8+3p)

Rich interface protocols

Supports interface protocols such as POSIX (SMB, NFS), Object, and HDFS:

• NFS v4.0, NFS v3

• SMB 3.0, SMB 2.1, SMB 2

• OpenStack Swift along with Keystone v3

• S3

• HDFS transparency 3.1.0-X, 3.0.0-X, 2.7.3-X, 2.7.2-X, 2.7.0-X

Scenario

Scenario Overview

Autonomous driving model training

For the large number of small files collected by devices such as in-vehicle cameras, radar, and infrared sensors in autonomous driving scenarios, CPFS Parallel File Storage provides low-latency, high input/output operations per second (IOPS) access. This can increase model training speed by more than three times.

Genome sequencing data assembly

Genome sequence assembly requires a massive number of concurrent computing jobs. CPFS Parallel File Storage provides access bandwidth of up to hundreds of gigabytes per second, meeting the simultaneous access demands of hundreds or thousands of nodes. This eliminates file I/O bottlenecks and reduces task completion time by 50%.

Oil exploration business analysis

Large amounts of geological data must be computed, processed, and analyzed. The raw seismic data and process data also require long-term storage. CPFS Parallel File Storage provides a single namespace at the petabyte (PB) level. It supports flexible management of space quotas and resource partitioning for different computing tasks. This meets the need for business expansion at any time.