Brendan Gregg's performance analysis methodology guide: USE, Off-CPU, Flame Graphs and more
DanielLockyer · x · 2026-10-02
- Brendan Gregg's classic Performance Analysis Methodology page resurfaces: a procedure for going from starting point to root cause in system/application performance work.
- Methods covered: Problem Statement, Workload Characterization, USE Method (resource bottlenecks), CPU Profile, Off-CPU Analysis (thread wait latency), Time Division, TSA Method, Active Benchmarking.
- Tooling: Flame Graphs, Flame Scope, Heat Maps, frequency trails, DTrace/eBPF tools.
- Key advice: analysis without a methodology becomes a fishing expedition; in complex environments, try several methodologies for about a week each before giving up, rather than sinking six months into one issue.
More from Infra
- PyTorch ships TorchTPU: vLLM and SGLang now run natively on Google TPUs — PyTorch · 2026-10-02
- Ethereum Foundation and Open Anonymity launch zkAPI: pay for AI APIs with zero-knowledge proofs — kenziyuliu · 2026-10-02
- Musk: SpaceX version of VR72 may run at close to 250kW average power with Nvidia — McDonaghMatthew · 2026-10-02
- DDR5/PCIe5 is 15-20% faster for LLM pre-training, but DDR4 wins on cost per dollar — Any-Winter-4079 · 2026-10-02
- PyTorch unveils TorchTPU backend so vLLM and SGLang run natively on TPUs — PyTorch · 2026-10-02
- Modal makes GPU clusters generally available with one-line RDMA support — josh_wills · 2026-10-02