Technical Overview of JD's Archimedes Resource Scheduling System

Introduction: The Archimedes project, awarded the 7th JD Technology Gold Award Special Team Contribution, is a resource scheduling system that evolved from JDOS 2.0, which introduced Kubernetes to large‑scale production in 2016.

From 2017, Archimedes took over all application resource scheduling at JD, tripling average cluster resource utilization and saving billions of yuan in server procurement during major sales events.

In 2018, the system increased business release efficiency by 10%, enabling resource requests in 0.2 seconds and delivering 10,000 containers within 10 minutes, thereby streamlining development and operations.

Application scenarios include rapid agile development and deployment, micro‑service architecture migration, and automatic elastic scaling for high‑traffic events such as flash sales.

Key challenges addressed were the limitations of theoretical scheduling algorithms; Archimedes combined algorithmic advances with practical engineering to achieve high‑throughput scheduling, resource monetization, mixed resource pools, and integration of big‑data and AI training workloads.

Business benefits comprise fine‑grained capacity planning, multi‑dimensional resource billing, and a 0.2‑second resource request time that enables second‑level scaling for online services.

Future development will focus on further improving resource time‑space reuse, fragment consolidation, fine‑grained operations, additional compute capacity, ultra‑large‑scale cluster construction, and cross‑data‑center scheduling.

The JD Technology Gold Award recognizes outstanding R&D teams and individuals, highlighting Archimedes as a flagship example of turning scheduling theory into engineering practice.