DapuStor has partnered with Pliops to provide a new alternative to the industry standard RocksDB storage engine, which is known to more than double performance and significantly improve the quality of service.
KVrocks, which is central to the critical operations of notable tech firms including Baidu, Trip.com Group, Meitu, RGYUN, U-NEXT, and Baishan Cloud, is an open-source, distributed key-value NoSQL database. It is designed to handle and process large-scale datasets and acts as a durable alternative to Redis by employing RocksDB for storage, ensuring compatibility with the Redis protocol. This persistent key-value store allows for significant memory cost savings when managing extensive datasets.
However, when data volumes increase to petabyte levels, meeting customer Service Level Agreements (SLAs) can become challenging. This difficulty is largely attributed to the performance and latency constraints of the RocksDB storage engine. Improving infrastructure and boosting operational efficiency are crucial measures to address these challenges, as they can reduce data storage costs while enhancing overall customer experience by ensuring responsiveness and reliability.
The DapuStor and Pliops collaboration aims to enhance KVrocks with superior key-value storage. The integration of DapuStor's SSDs with Pliops' AccelKV is expected to offer substantial benefits for KVrocks users.
DapuStor specialises in enterprise and data centre SSD storage solutions, supported by its strong portfolio in ASIC controller chip technology. Grant Li, DapuStor's solution architect, explained, "The DapuStor R5101 is a high-performance PCIe4.0 NVMe SSD that offers superior reliability, low latency, and excellent power efficiency, optimising TCO for enterprise IT and cloud facilities. Ideal for core data storage across various sectors like logistics, finance, intelligent manufacturing, and AI, it achieves 1,750 randread and 240 randwrite IOPS. Combined with Pliops XDP and library, the solution delivers a high-performance key-value storage interface."
Pliops' XDP is a PCIe-based accelerator designed for seamless integration with servers and SSDs, ensuring swift deployment and operation. Its XDP-Rocks library extends RocksDB functionalities, optimising KVrocks integration with features such as efficient data manipulation and advanced dataset scanning, which are crucial for robust data retrieval.
Tony Afshary, Global VP of Products and Marketing at Pliops, stated, "At Pliops, we lead the industry in providing efficient and intelligent data transport solutions. Leveraging our innovative Key-Value technology, our products are designed to accelerate critical AI applications on both GPU and CPU platforms. We're thrilled to partner with DapuStore and accelerate their best-in-class NVMe storage solutions. Our customers will be delighted when deploying these integrated products."
The partnership's effectiveness was demonstrated using the Dbbench database benchmark tool to compare Pliops AccelKV's performance for KVrocks against RocksDB. The benchmark involved a dataset of 1.2 billion objects and revealed that Pliops KV outperformed RocksDB by a factor of two, demonstrating double the performance. Furthermore, it offers enhanced scalability and improved system efficiency for KVrocks applications.
The combination of DapuStor R5101 SSDs with Pliops XDP AccelKV proved instrumental in pushing the systems to their maximum capacities during testing. The results indicated that Pliops KV is a viable and superior alternative to conventional RocksDB implementations across various applications, especially KVrocks.
One notable advantage of Pliops KV is its significant impact on Quality of Service (QoS). In mixed workloads involving 50% read and 50% write operations, tail latency dropped between 66% and 87%. For workloads with 70% read and 30% write operations, the reduction ranged between 10% to 28%. This substantial reduction in latency ensures swift and reliable data access, which is crucial for delivering an exceptional customer experience.
By leveraging existing infrastructure and reducing the need for higher memory capacity, the solution offers a cost-effective option for large-scale data management. Grant Li of DapuStor noted, "This solution brings a 53% reduction in price performance due to performance acceleration."
