Alibaba's In-Storage Computing SSD (AliFlash V5) and the Evolution of AliFlash Products

Alibaba announced at the 2019 Flash Memory Summit a next‑generation integrated compute storage architecture for data centers – the Alibaba In‑Storage Computing SSD (ISC SSD), also known as near‑storage computing SSD. This follows the 2018 open‑interface SSD announcements at FAST and FMS, marking another heavyweight innovation in storage technology.

Alibaba's self‑developed AliFlash SSD series has evolved through multiple generations. The first generation, AliFlash V1, was a host‑based PCIe SSD that achieved great success with tens of thousands of online deployments. The second generation, AliFlash, became a standard NVMe SSD with wide deployment. The third generation, AliFlash V3, introduced an open‑channel interface and Alibaba released its own Open‑Channel (AOC) specification, establishing a thriving ecosystem with major SSD vendors; V3 is now in mass production. The latest generation, AliFlash V5, implements the near‑storage computing architecture and is currently in small‑scale trial.

Evolution of Alibaba's self‑developed AliFlash products

AliFlash V5 continues to use the open‑interface features of AliFlash V3, exposing an open Flash Translation Layer (FTL) interface that fuses FTL with upper‑layer software, reducing write amplification and I/O latency. AliFlash V5 creatively enables near‑storage computing, allowing data to be processed close to storage, greatly improving compute efficiency and lowering system latency. This transformative innovation supports the evolution of next‑generation data‑center storage architectures.

Alibaba's presentation at the 2019 Flash Memory Summit

Data centers are increasingly adopting a storage‑compute separation architecture, but transferring data between storage and compute nodes wastes network bandwidth and CPU resources, increasing system latency. Alibaba's near‑storage computing (In‑storage computing) places compute capability inside the storage component, reducing CPU and memory load on compute nodes, alleviating network load, and significantly lowering application response latency, thereby providing new solutions for big‑data and AI workloads.

Business validation shows notable benefits. In a relational‑database scenario, using traditional compute‑node CPUs for basic database operators (decompression, projection, filter) becomes a bottleneck under high load, causing query latency to rise.

When using In‑storage computing SSD offload, computation runs on the SSD side, dramatically reducing CPU burden and network latency. Even under equal or greater load, the system maintains low latency, achieving roughly a 3× bandwidth increase and an 88% latency reduction compared to CPU‑only testing.

Alibaba continues to build a technology ecosystem, establishing extensive industry collaborations. The ISC (In‑storage Computing) project has received broad support and its ecosystem is continuously expanding.

Ecology of In‑storage Computing SSD announced at the 2019 Flash Memory Summit

Alibaba's In‑storage Computing SSD ecosystem has demonstrated clear benefits in data applications. Alibaba will continue to productize In‑Storage Computing SSDs and expand cooperation with industry partners to broaden near‑storage computing scenarios in data centers, creating greater value for AI, cloud computing, and related services.