On February 27 at USENIX FAST 2026 in Santa Clara, the ApsaraDB PolarStore paper “PolarStore: High — Performance Data Compression for Large — Scale Cloud — Native Databases” was selected as one of five Best Paper candidates. The paper was chosen from 44 accepted submissions at a conference that reported an overall acceptance rate of 17.4%. The submission credits a single contributor organization and presents a production — ready approach to compression for cloud — native databases, a development that could affect storage efficiency and performance for large cloud deployments.
The research describes a multilayer, software — hardware compression architecture that aligns database I/O characteristics with storage hardware features. Its design aims to eliminate redundant compression across software and hardware layers, reconcile high throughput with high compression ratios, and enable flexible scheduling of compression tasks between layers. The authors report this design has been implemented commercially, claiming an industry — first software — hardware collaborative compression product deployed at large scale in production cloud — native databases.
PolarStore is presented as the core storage foundation for PolarDB’s storage — compute separation architecture. Implemented as a distributed storage system engineered for ultra — low latency and high availability, PolarStore uses a lightweight user-space network and I/O stack that bypasses the traditional kernel path to better leverage emerging hardware capabilities and reduce I/O overhead.
From an engineering perspective, the paper emphasizes a hardware — aware stack that explicitly leverages RDMA and NVMe SSDs to reduce end-to-end latency for distributed non-volatile data access. The combination of software — hardware compression with a user-space I/O approach is claimed to deliver the performance characteristics required for both cloud — native transactional and analytical workloads, and the authors state the system is already applied at production scale within their database environment.
PolarDB for MySQL/PostgreSQL is described as the cloud — native database that uses PolarStore as its storage layer. The system is reported to be fully compatible with MySQL and PostgreSQL, to support HTAP workloads, multi — master multi — write configurations, multi — active disaster recovery, and to include built — in AI capabilities. The company cites transaction and analytical performance advantages of up to 2–3× and 400× versus open-source databases respectively, and estimates a total cost of ownership reduction of roughly 50% compared with self-managed databases.
FAST is organized by USENIX and ACM SIGOPS and has run since 2002 as a leading venue for file and storage technologies. In naming the paper a Best Paper candidate, the judges highlighted that the work systematically abstracts existing compression techniques, pinpoints industry pain points, and contributes large — scale, long-term engineering experience back to the research community — qualities that underline its relevance to both academic researchers and industrial practitioners.
Sources
Replies (0)
No replies in this topic yet.
