Anthropic为应对Claude自主代码生成带来的开发效率瓶颈，设计并构建了名为Scion的自主版本控制系统1。该公司在2026年5月至9月期间，每日合并的代码变更从2,500次增长至30,000次1，为此需要重新架构其源代码管理基础设施。一支七人团队用五个月时间完成了从Git到Scion的全面迁移，同期还开发了Antique代码审查工具1。
Scion在基础设施成本和性能上实现了显著提升。相比Git镜像方案，新系统的CPU和内存成本降低了200倍1。代码检出速度获得100倍的性能提升，使CI和研究任务的构建时间从分钟级降至秒级，同时为CI工作流节省了25%的CPU容量1。迁移涉及近100GB的Git历史记录和超25GB的压缩副本数据，该系统每日需处理2,500万次代码检出操作1。系统采用服务器-客户端架构设计，每个缓存层维持99%以上的缓存命中率1。此外，使用Claude Mythos Preview处理手写的Git-Scion摄入管道后，性能又提升了20倍，内存消耗同时降低4倍1。
Anthropic has developed and deployed Scion, a proprietary version control system designed to handle the infrastructure demands created by Claude's autonomous code generation capabilities.1 Between May and September 2026, the company experienced explosive growth in development velocity, with daily merged code changes increasing from 2,500 to 30,000.1 A seven-person team completed both the development of Scion and the Antique code review tool alongside a full migration from Git within five months.1
The new system demonstrates substantial performance improvements over the previous Git-based infrastructure. Scion reduced infrastructure costs by 200 times compared to mirrored Git deployments in terms of CPU and memory consumption.1 Code checkout speeds for CI and research tasks improved by 100 times, dropping from minutes to seconds, while simultaneously achieving a 25 percent CPU capacity gain for CI operations.1 The system maintains cache hit rates exceeding 99 percent at each caching layer.1 Additionally, Claude Mythos Preview integration with the hand-written Git-Scion ingestion pipeline delivered a 20 times performance boost and reduced memory consumption by 4 times.1
The migration handled substantial data volumes, with Git history comprising nearly 100 gigabytes and its compressed copy exceeding 25 gigabytes, while supporting 25 million daily code checkouts.1 The implementation reflects a broader approach to infrastructure optimization powered by AI-assisted systems engineering at the company.
评论
还没有评论，欢迎留下第一条。