From Open‑Source to Self‑Developed: JD.com’s One‑Month Migration of Hundreds of Millions of Images

Images have been a core component of JD.com since the launch of its e‑commerce platform, growing from a few thousand to over a hundred million new images added daily, reflecting the rapid expansion of the business.

In the early years, open‑source software drove the rapid development of JD’s image system, but by 2014 the data volume reached a tipping point, prompting a shift to a fully self‑developed distributed file system called JFS. The migration had to be completed within a month, coinciding with JD’s IPO.

The newly built image platform delivered significantly higher performance and response speed, handling traffic spikes during major sales events (e.g., 618, Double‑11) without degradation, and continues to scale as daily image volume grows by billions.

To address image compression, JD partnered with Intel to optimize the compiler layer, achieving more than a two‑fold performance boost over the previously dominant Graphic Magic (GM) solution.

JD also adopted Google’s WebP format early on, reducing average image size by 40‑55% compared to JPEG, which saved bandwidth and accelerated download speeds, especially for mobile users.

Leveraging recent advances in neural networks, JD experimented with AI‑based image compression, achieving further size reductions without compromising user experience.

The article then introduces “blind watermark” technology, which embeds invisible watermarks using wavelet decomposition, random number generation, and redundant coding, enabling detection of unauthorized distribution while remaining robust against cropping, format changes, and other alterations.

JD’s blind‑watermark implementation uses a parallel loop design for random number generation and watermark embedding/decoding, providing optimized performance.

Looking ahead, JD plans to continue innovating in image processing and security, promising new “black‑tech” developments beyond blind watermarking.