Alessio Toniolo

I'm a CS student at CMU and Member of Technical Staff Intern at Modal Labs.

Alessio Toniolo

Writing

RDMA in gVisor: A Deep Dive into GPU Networking

RDMA enables high-bandwidth communication between machines in an HPC cluster. It has been a long-requested feature in gVisor, a popular container runtime. This post walks through the internals of RDMA and how we added support for it to gVisor.

GORGO: Online Tuning for Cross-Region Network-Aware LLM Serving

Existing cache-aware LLM routing policies struggle to balance load effectively. GORGO jointly optimizes network latency, prefill time, queueing delay, and KV-cache reuse, reducing TTFT and E2E latency by 6–30% over policies such as session affinity.