Alibaba Cloud Showcases ALS System and AI‑Driven Storage Innovations at CCF China Storage Conference 2024

From November 29 to December 1, the CCF China Storage Conference was held at the Guangzhou Chimelong International Convention & Exhibition Center, where Alibaba Cloud’s server research team delivered talks on storage innovation in the super‑computing internet era and AI‑driven scenarios, and announced the current status of the ALS (ALink System) ecosystem.

Background: The annual CCF China Storage Conference gathers academia and industry to explore over twenty technical fields, including new memory devices, storage interconnects, distributed computing, databases, file systems, and AI‑centric storage solutions, providing a platform for experts, scholars, and practitioners.

The UALink Alliance, founded on October 29, quickly grew to nearly 40 members, including major CSPs, indicating a consensus on Scale‑Up interconnect technology for massive AI compute infrastructure. Alibaba Cloud, together with the China Academy of Information and Communications Technology and AMD, released the ALink System ecosystem and demonstrated the AI Infra 2.0 server system, whose low‑level interconnect protocol is compatible with UALink.

Scale‑Up vs. Scale‑Out: According to Alibaba Cloud’s high‑speed interconnect lead Kong Yang, Scale‑Up protocols (e.g., UALink) connect GPUs directly, while Scale‑Out protocols (e.g., UEC) interconnect servers via NICs and switches; the two differ in scope, traffic patterns, and programming models, influencing protocol design and hardware implementation.

ALS (ALink System): ALS is an open ecosystem led by Alibaba Cloud to standardize Scale‑Up interconnect solutions. It consists of ALS‑D (data plane) using the UALink protocol with high bandwidth, ultra‑low latency, memory‑semantic access, and in‑network computing features, and ALS‑M (management plane) offering standardized chip access, multi‑tenant configurations, and flexible integration for both open and proprietary protocols.

AI Storage Innovations: Senior technical expert Tao Kai presented Alibaba Cloud’s multi‑layer storage innovations for AI, ranging from component‑level designs to server‑level solutions. These include hierarchical storage for large models, intelligent data‑flow platforms, high‑bandwidth/low‑latency storage servers, large‑capacity QLC SSDs, and high‑bandwidth/low‑latency CXL persistent memory.

During the development of self‑designed components Aliflash and AliSCM, Alibaba Cloud continuously optimizes cost, operational efficiency, and deep hardware‑software integration to address business pain points and enable customized solutions, while exploring CXL, PIM, and other emerging technologies.

On the server side, Alibaba Cloud builds on the PanJiu storage server platform to deliver high‑performance servers with 400 Gbps low‑latency networking, large‑capacity SMR HDD solutions for massive AI datasets, and AI‑focused storage servers that support intelligent data movement, acceleration, and workflow optimization.

Alibaba Cloud’s server R&D remains driven by technological innovation, aiming to lead industry trends through deep software‑hardware convergence and to create unique value for its infrastructure ecosystem.