Domestic ZHaoxin KHS-40000 CPU Expandable to 262,000 Cores! Hard Drive Exceeds 8.8 Million TB

On April 9th, domestic x86 CPU processor manufacturer ZHaoxin announced, in collaboration with Phytium Information, the launch of the high-performance massive distributed storage solution "UbiScale 12000" based on the ZHaoxin KHS-40000 series processor platform. Designed for massive unstructured data storage scenarios, it boasts nearly limitless scalability. The scheme supports up to 4096 nodes, each capable of hosting one or two Megachips KH-40000/16 16-core or KH-40000/32 32-core processors, totaling a maximum of 262,144 cores.

Nodes come in heights of 2U or 4U, with each node able to accommodate up to 90 hard disk drives (HDDs). Considering the current maximum capacity of 24TB per drive, the total capacity can exceed 8EB, precisely 8,847,360TB, or more than 8.8 million TB.

Additionally, each node has two SATA SSDs for the system disk, with a maximum memory of 4TB. With 4096 nodes, that's a total of 16PB. The scheme employs a fully symmetric, decentralized distributed cluster architecture, combined with large-scale EC (Erasure Coding) technology, to integrate large-capacity hard drives into a unified resource pool, providing high-reliability, low-cost storage space for upper-layer applications.

Storage nodes are interconnected with dual interaction networking, with each node having the same role and no dedicated metadata service design, eliminating metadata bottlenecks. Capacity performance scales linearly with the expansion of the cluster size, with data automatically balanced and distributed, and automatic load balancing between nodes.

Features:

1. Powerful Functionality

   • Decentralized Architecture: Fully symmetric distributed architecture, no centralized metadata design, eliminating performance bottlenecks.
   • Unlimited Scalability: Supports over 4096 nodes, nearly limitless scalability.
   • Ultra-Large Capacity: Single cluster with a capacity of over 8EB.
   • Unified Namespace: Provides a unified namespace to the outside world.

2. Excellent Performance

   • Linear Performance Growth: Distributed symmetric architecture, performance grows linearly with the addition of nodes.
   • Unstructured Data Storage: Massive image storage and retrieval within seconds.
   • High-Performance Read/Write: Single-node bandwidth can reach up to 5.0GB/s.

3. Stability and Reliability

   • Inter-cluster Reliability: Supports replication across geographical clusters, with rapid switching between primary and backup sites.
   • Cluster-level Reliability: Decentralized fully symmetric architecture ensures that the failure of any node does not affect business operations.
   • Object/File-level Reliability: EC redundancy encoding, N+M data redundancy protection, with N+M up to a maximum of 64. The recommended range for M is 2 to 8, supporting the tolerance of data loss from 8 node failures without interruption of service.

Example Use Cases:

1. Video Surveillance Applications
   • Provides nearly unlimited storage capacity and performance scalability.
   • Supports a streaming storage architecture, compatible with mainstream video protocols such as GB/T28181 and Onvif. Each node supports the access of 1600 streams at 4Mbps, with a maximum disk utilization of 96.88%.
   • Supports intelligent video repair to maximize protection of video data.
2. Applications in Medical Research

Significantly improve the performance of PACS small file read and write operations through technologies such as intelligent grading, intelligent caching, and intelligent aggregation.

Provide long-term secure storage capacity while complying with regulatory requirements, meeting the demand for fast retrieval of original image data anytime, anywhere. 3. Application Scenarios in the Financial Industry

For unstructured data, intelligent retrieval, caching, classification, and aggregation techniques are provided for performance optimization. This enables storage and retrieval of large quantities of small images in milliseconds. Media Asset Use Cases

As a unified storage pool for media assets, it enables dynamic sharing of data for acquisition, editing, on-demand viewing, management, and storage business processes.

A single cluster can achieve storage capacities of up to 7.4EB, meeting the demands for storage capacity of 4K and 8K ultra-high-definition resources.

With a decentralized fully symmetric distributed architecture and hardware redundancy design within nodes, it ensures the long-term reliable operation of business processes. Currently, there are several server products available featuring the massive storage solution from Zhaoxin and Phytium Information.

For instance, the Lenovo Tianqi KR722z G2 is a 2U general rack-mountable server with front support for 12 3.5-inch or 24 2.5-inch hot-swappable hard drives. It also has rear support for 4 hot-swappable hard drives, with a maximum memory capacity of 2TB. The entire machine adopts redundant cooling and optional power supply design. For example, the SuperCloud R3210 Z11 is equipped with dual Megachips Opensilicon KH-40000/32 processors, supporting up to 12 front-accessible 3.5-inch or 24 hot-swappable 2.5-inch hard drives, as well as 2 rear-accessible hot-swappable 2.5-inch hard drives. It can accommodate a maximum of 32 memory modules and 6 PCIe slots.