Posts

A technical note on GPU checkpoint/restore for schedulers, using FlowGPU as the main reference and my cudaw prototype as the first version of the codebase.

A technical guide to GPU datacenter scheduling based on our ACM Computing Surveys paper, covering training, inference, HPO, mixed workloads, and future scheduler design.

A technical report on ResiHP, a resilient training system that detects fail-slow devices under noisy sequence-length variation and dynamically adapts 3D parallelism.

A technical note on CONCUR, an agent-level admission control layer that prevents KV cache collapse during long-running agentic LLM inference.

A technical note on ASTRAEA, a multi-tenant GPU scheduler that measures fairness by long-term GPU-time instead of instantaneous allocation or finish time alone.

A technical story behind Hydro’s Bubble Squeezer, which runs surrogate hyperparameter tuning trials inside the idle bubbles of pipeline-parallel large-model training.

Serveral optimizations and current services related to this site.