Blog

Quickly get started with openYuanrong and start your distributed computing journey! This article will take you through the core concepts and basic usage of openYuanrong, helping you get started quickly through simple examples.

In-depth analysis of openYuanrong's core technical concepts, understanding how to simplify complex distributed clusters into a standalone programming experience.

Continue to explore openYuanrong's core architecture design, revealing how it achieves high-performance distributed computing.

As a general-purpose Serverless distributed computing engine, openYuanrong fully considers multi-tenant isolation from the beginning. It supports multiple tenants sharing the same cluster, while deploying FunctionProxy and DataWorker components as DaemonSets, achieving node-level resource sharing and effectively accelerating cold start efficiency for tenant workloads.

Understand in depth how openYuanrong implements dynamic function scheduling in large-scale distributed environments, supporting rapid elastic scaling and cross-node migration of function instances.

Learn how openYuanrong data system achieves high-performance data sharing through heterogeneous distributed multi-level caching.

Explore how openYuanrong data stream eliminates redundant copying and transfer overhead in traditional message middleware.

Learn in depth how openYuanrong serves as the KV backend for TransferQueue, relying on distributed heterogeneous object multi-level caching capabilities oriented to the Ascend platform, accelerating veRL's sample transfer link, achieving 3-4 times improvement in data transfer efficiency and 40% improvement in RL end-to-end performance.

As AI Agent applications move from personal assistants to enterprise-grade production deployment, how to enable JiuwenSwarm to break through single-machine limitations and achieve enterprise-grade elasticity and reliability while maintaining the existing programming experience is the core engineering problem that openYuanrong aims to solve. This article explores how to perform distributed transformation of JiuwenSwarm based on openYuanrong, achieving a smooth evolution from single-machine deployment to cluster-based operation for AI Agents.

The non-determinism of Agent applications brings unique operational characteristics and challenges such as high dynamism, insecurity, and long sessions, which traditional K8s container microservice technology systems struggle to meet. openYuanrong, with its distributed kernel philosophy similar to a standalone OS, naturally supports Agent instance auto-elasticity, secure sandbox isolation, session-affinity scheduling, and distributed fault tolerance, providing matching distributed infrastructure for enterprise-grade large-scale Agent applications.

openJiuwen and openYuanrong have bridged the full chain between Agent Runtime, openYuanrong, RL training framework (verl), and Ascend computing power. Each real execution trajectory produced by an Agent is structured as learnable samples, directly entering the GRPO training loop, completing inference and training on Ascend environments, achieving an automated closed-loop path from "execution trajectory" to "policy update".

We design, implement, and evaluate YuanRong, the first production general-purpose serverless platform with a unified programming interface, multi-language runtime, and a distributed computing kernel for cloud-based applications.