DeepSeek Elastic Compute (DSec)

Abstract: Large-scale agentic training and evaluation with large language models (LLMs) rely on isolated, stateful execution environments in which models inspect repositories, invoke tools, execute commands, and interact with task-specific services. These workloads create sandboxes in large bursts, span heterogeneous functionality and isolation requirements, retain state across long interactions, and draw from large image corpora with limited reuse. Supporting them therefore requires an elastic execution platform rather than a single sandbox runtime.

摘要： 基于大语言模型（LLM）的大规模智能体（Agentic）训练与评估，依赖于隔离且有状态的执行环境，模型在其中检查代码库、调用工具、执行命令并与特定任务服务进行交互。这些工作负载会产生大规模的沙箱突发需求，涵盖异构的功能与隔离要求，在长交互过程中保持状态，并从重用率有限的大型镜像库中提取数据。因此，支持这些需求需要一个弹性的执行平台，而非单一的沙箱运行时。

This report presents DeepSeek Elastic Compute (DSec), a production sandbox platform that exposes FnCall, container, microVM, and full-VM sandbox backends through a unified SDK. DSec coordinates placement and lifecycle management across the cluster, composes environments from independently versioned layers, combines memory sharing, reclamation, and CPU scheduling for high-density execution, and loads image data on demand from Fire-Flyer File System (3FS), a cluster-wide distributed filesystem.

本报告介绍了 DeepSeek Elastic Compute (DSec)，这是一个生产级沙箱平台，通过统一的 SDK 暴露了 FnCall、容器、微虚拟机（microVM）和全虚拟机（full-VM）沙箱后端。DSec 在集群范围内协调部署与生命周期管理，通过独立版本化的层构建环境，结合内存共享、回收和 CPU 调度以实现高密度执行，并从集群范围的分布式文件系统 Fire-Flyer File System (3FS) 中按需加载镜像数据。

DSec is co-designed with the reinforcement learning (RL) framework, decouples stateful rollout execution from preemptible GPU training, coordinates sandbox lifecycle with training to preserve rollout state while reclaiming idle resources, and mitigates agent misbehavior such as reward hacking. A single production-scale unit of DSec spans around 160 nodes, serving about 3 million sandboxes per day; in production, it supports over 380,000 concurrent sandboxes and sustains over 5,000 sandbox creations per second.

DSec 与强化学习（RL）框架协同设计，将有状态的 Rollout 执行与可抢占的 GPU 训练解耦，协调沙箱生命周期与训练过程，在保留 Rollout 状态的同时回收空闲资源，并减轻智能体如奖励作弊（Reward Hacking）等不良行为。DSec 的单个生产规模单元跨越约 160 个节点，每天服务约 300 万个沙箱；在生产环境中，它支持超过 38 万个并发沙箱，并维持每秒超过 5,000 个沙箱的创建速度。

Our evaluation and deployment experience show that these mechanisms reduce environment setup and image-distribution overhead, improve memory efficiency, and preserve latency-sensitive performance under high-density overcommit.

我们的评估和部署经验表明，这些机制降低了环境设置和镜像分发的开销，提高了内存效率，并在高密度超卖（Overcommit）情况下保持了对延迟敏感的性能。