Interview with JD.com Infrastructure Director Bao Yongcheng on Container Platforms, Kernel Optimizations, and Scaling for the 11.11 Summit

Q: What research and development work have you been responsible for after joining JD.com?

A: Since 2013 I have been focused on containers, which act as the runtime platform and encompass a wide range of supporting systems such as data‑center management, kernel, operating system, logging, monitoring, Docker, OpenStack, distributed storage, and the complete business‑on‑boarding workflow.

Q: What is the focus of your container research and what results have you achieved?

A: The core focus is the Linux kernel. While Docker and OpenStack are management tools, the kernel itself determines container performance and stability; container technology is fundamentally Linux‑kernel technology.

Q: What kernel‑related achievements have you made?

A: For the "11·11 Global Good Products Festival" we upgraded the CPU scheduling algorithm in the kernel, enabling seamless dynamic allocation of infrastructure resources. This allowed every thousand servers to deliver the compute power of two thousand servers, dramatically reducing procurement and O&M costs and raising average CPU utilization to industry‑leading levels.

We also instrumented the kernel with numerous hooks and probes to monitor all container‑level behavior, allowing us to pinpoint performance regressions at the instruction‑set level and provide developers with precise diagnostics.

Q: Did anything interesting happen during the R&D work for the 11·11 promotion?

A: By automatically throttling CPU frequency based on usage, I/O, and workload metrics across two physical PODs, we saved about 17% of data‑center power consumption. The technique will be rolled out to all JD.com data‑centers in 2018 to further reduce energy use.

Q: How do you view Kubernetes and OpenStack?

A: OpenStack is mature but somewhat heavyweight, whereas Kubernetes is at an early, rapidly evolving stage similar to OpenStack’s first two years. Kubernetes requires deep integration and custom development to handle large‑scale deployments; JD.com has heavily customized many components to suit its massive workload.

When clusters reach 1,000, 3,000, or 6,000 nodes, we encounter specific scaling challenges, and I will share those experiences at the upcoming JD.com technology summit.

Q: What are your next development priorities?

A: We aim to roll out the Archimedes platform across all JD.com data‑centers, integrating it into every container cluster to ensure stable, high‑performance operation while cutting total cost of ownership and halving server purchases by 2018. We also plan to open‑source mature internal products to the community.

Q: How do you see the future trend of the industry?

A: Since JD.com began using containers in 2014, by early 2016 100% of our services ran in containers, and by last year over 95% of databases were containerized. We believe containers will become the dominant model for both public and private clouds, leading to an intelligent, container‑centric data‑center future.

Q: How many major promotions have you participated in, and what has changed?

A: I have taken part in ten major promotions. The constant is the technical focus of the preparation; the change is the continual innovation of the technology stack, requiring both rapid adoption of new trends and proactive research.

Q: What will you share at the upcoming JD.com "11·11" infrastructure summit?

A: I will present JDOS Archimedes development experiences, how we handle urgent production issues without affecting online services, and how the platform improves server utilization, enabling more business to run on fewer machines. I will also discuss the full rollout of JDOS 2.0, built on Kubernetes, and its role in supporting massive promotional traffic.