Alibaba Cloud PolarDB PolarStore Receives Best Paper Award Nomination at Top Conference FAST'26

This article introduces Alibaba Cloud PolarStore's software-hardware collaborative compression architecture for cloud-native databases, which received a Best Paper Award nomination at USENIX FAST'26.

On February 27, at the top conference in the storage realm held in Santa Clara, USA, USENIX FAST 2026, the Alibaba Cloud ApsaraDB PolarStore team's paper "PolarStore: High-Performance Data Compression for Large-Scale Cloud-Native Databases" won the Best Paper Award Candidate. This achievement stood out from 44 accepted papers globally (acceptance rate 17.4%) and became one of only 5 best paper candidates.

Alibaba Cloud database innovatively designed a software-hardware multilayer compression architecture, deeply integrating the I/O features of cloud-native databases. This breaks the semantic barriers between databases and storage, eliminates software and hardware compression feature redundancy, and achieves a unification of high performance, high compression ratio, and flexible scheduling. Based on this technology, Alibaba Cloud built the industry's first commercially implemented software-hardware collaborative compression product, which has been deployed and applied on a large scale in the cloud-native database production environment. This paper, with Alibaba Cloud as the sole contributor, systematically summarizes the large-scale engineering practice and technical insights of software-hardware collaborative compression, setting a new benchmark for cloud database storage optimization.

The FAST conference (USENIX Conference on File and Storage Technologies) is jointly organized by the Advanced Computing Systems Association (USENIX) and the Association for Computing Machinery Special Interest Group on Operating Systems (ACM SIGOPS). Founded in 2002, it is a recognized international top academic conference in the storage realm and is listed as a Class A conference by the China Computer Federation (CCF).

The conference focuses on frontier directions such as file systems, cloud storage, new storage hardware, and data management. The paper acceptance standards are strict, and the results represent the highest research level in the industry, highly valued by the global academic and industrial communities. For more than 20 years since its inception, FAST has continuously promoted the development and breakthroughs of key technologies such as software-hardware collaboration, RAID, flash memory file systems, non-volatile memory technology, and distributed storage. This has had a profound impact on industrial practice.

PolarDB for MySQL/PostgreSQL is a cloud-native database self-developed by Alibaba Cloud. It adopts an integrated software-hardware design, is based on the high-performance storage PolarStore, and is 100% compatible with MySQL/PostgreSQL. It supports hybrid transactional and analytical processing (HTAP), multi-master multi-write, multi-active disaster recovery, and built-in AI. The transaction and analysis performance are up to 2 to 3 times and 400 times that of open source databases, respectively, and the total cost of ownership (TCO) is 50% lower than that of self-managed databases.

As the core foundation of the PolarDB storage-compute separation architecture, PolarStore is a distributed storage system with ultra-low latency and high availability capabilities, which is crucial for leveraging PolarDB attributes. PolarStore adopts a lightweight user space network and I/O stack, bypassing the traditional kernel stack. It fully leverages the potential of emerging hardware such as Remote Direct Memory Access (RDMA) and Non-Volatile Memory Express (NVMe) solid-state drives (SSDs), greatly reducing the end-to-end latency of distributed non-volatile data access.

The FAST judges commented: "By systematically abstracting existing compression technologies and providing precise insights into industry pain points, the paper innovatively built a software-hardware collaborative multilayer compression architecture. At the same time, it deeply contributes large-scale, long-term stable engineering practices back to the academic community. This provides a reference paradigm with both theoretical depth and practical value for research directions such as hardware architecture design, compression algorithm optimization, and intelligent scheduling policies, highlighting the important leading role of frontier industrial practices in academic innovation."

The software-hardware collaborative compression scheme proposed in the paper achieves full-link deep collaboration for PolarDB: The software layer deeply perceives InnoDB Page borders, compresses with 16 KB large block granularity, and performs full-link I/O tagging. The hardware layer relies on Smart-SSD variable-length Flash Translation Layer (FTL) capabilities to compactly organize compressed blocks. This fuses software and hardware logic and completely eliminates the multilayer Garbage Collection (GC) and I/O amplification problems in traditional schemes. Online practice verifies that this scheme achieves an average compression ratio of 3.55 times, with a maximum instance-level compression ratio exceeding 10 times. It also features zero I/O amplification and requires no garbage collection.

In terms of performance, under the premise of 100% compatibility with the InnoDB engine, only data pages are compressed, bypassing the Redo/binary logging path. CPU overhead is offloaded to storage, and the Redo link is efficient and lossless. Combined with EMP features, this achieves lossless performance or even performance improvements. At the same time, it supports the dynamic adjustment of compression algorithms, compression granularity, and single I/O policies. The software stack is concise with no background traffic, eliminating Performance Fluctuation.

As the industry's first commercialized software-hardware dual compression product, this solution significantly reduces ApsaraDB storage costs and Total Cost of Ownership (TCO). While ensuring extreme stability and business flexibility, it provides an efficient, economical, and reliable storage foundation for large-scale cloud-native applications.

Figure: Introduction to PolarStore software-hardware collaborative compression

The paper “PolarStore: High-Performance Data Compression for Large-Scale Cloud-Native Databases”, independently completed by the Alibaba Cloud ApsaraDB team, systematically summarizes the large-scale engineering practices and technical insights of software-hardware collaborative compression, setting a new benchmark for ApsaraDB storage optimization. It is reported that PolarDB has been applied on a large scale to the core operational systems of industries such as government affairs, finance, telecommunications, logistics, the Internet, and automotive, helping developers accelerate their entry into the Intelligent era.

Community

Alibaba Cloud PolarDB PolarStore Receives Best Paper Award Nomination at Top Conference FAST'26

Read previous post:

Read next post:

ApsaraDB

You may also like

Comments

ApsaraDB

Related Products

PolarDB for PostgreSQL

PolarDB for Xscale

PolarDB for MySQL

Database for FinTech Solution