CNware WinStore Distributed Storage – High Performance Edition
Distributed storage utilizes decentralized distributed technologies to virtualize the hard drives across multiple servers, thereby creating a unified virtual storage resource pool that provides external storage space. This approach enables flexible horizontal scaling of storage capacity and performance, while ensuring the security and reliability of stored data through redundancy strategies.
WinHong CNware WinStore Distributed Storage (High-Performance Edition) is a high-performance storage solution designed specifically for NVMe SSDs. Built on a next-generation distributed storage architecture, it delivers exceptional performance advantages in storage-computing separation scenarios, fully leverages the strengths of storage media such as all-flash and hybrid flash arrays, significantly reduces read/write latency, and meets the stringent performance requirements of mission-critical business applications.
Application Scenarios
Product Features
Globa l lock-f ree architectureBy leveraging the High-Speed Non-Lock I/O Stack, it eliminates line pool lock contention, enabling true end-to-end performance with zero context switches, zero-lock interrupts, and zero memory copying – allowing single-volume IOPS to reach the millions range.
Kernel BypassBy leveraging the advantages of the RDMA protocol and bypassing the traditional TCP protocol stack, this approach enables data transmission over the shortest possible path, achieving a single-link latency as low as tens of microseconds – a significant reduction compared to traditional SDS latency.
Multi-level caching algorithmBy leveraging technologies such as cache management, metadata management, small-batch data aggregation and disk flushing algorithms, large-batch sequential data transfer, elevator scheduling algorithm, inmemory cache, and pre-read algorithms, superior performance is achieved.
Multi-core NUMA-Aware OptimizationDeeply optimized NUMA architecture for domestic multi-core CPUs such as C86 and ARM, combined with lossless NUMA scheduling, fully unleashes the performance potential of domestic hardware.
Perfect Linear ExtensionThe distributed architecture supports horizontal scaling, with performance increasing linearly as the number of nodes grows; in practical tests on a 32-node cluster (4 K random read operations), the system achieved over 100 million I/O operations per second (IO PS).
Access Documents
7*24-hour Service Hotline
400-6300-003
