Is sglang's Scheduling Scope Designed for Cluster-Level Management (AIBrix + vllm Style)?

[CormickKneey]

Hi ~

I've been exploring sglang and noticed that its design incorporates features like radix-cache scheduling and kv-cache transfer(in pd disaggregation). These aspects suggest a robust caching and scheduling mechanism that might extend beyond single-node optimizations.

I'm curious: is sglang's scheduling scope intended to operate at a cluster level? In other words, is the project positioned to provide a solution similar to an AIBrix combined with vllm setup—managing high-throughput inference and caching efficiently across multiple nodes?

Any insights or clarifications regarding the long-term vision for distributed scheduling in sglang would be greatly appreciated. 😄😄

Thank you!