Speedscope

Speedscope
Name	Speedscope

Speedscope is an interactive open-source profile visualization tool for analyzing performance traces from software like Node.js, Google Chrome, Firefox, Visual Studio Code, Rust, and Go. It provides flamegraph and left-heavy visualizations suitable for investigating CPU hotspots in applications deployed on platforms such as Linux, macOS, and Microsoft Windows. Developers, engineers, and researchers from organizations including Google, Mozilla, Microsoft, Facebook, and Netflix use Speedscope in conjunction with profilers like Linux perf, DTrace, eBPF, perfetto, and Xcode to accelerate performance debugging and optimization workflows.

Overview

Speedscope presents performance traces as interactive visualizations including "icicle" flamegraphs and "left-heavy" stacks used by projects like Brendan Gregg's tooling, Flame Graph collections, and Brendan Eich-era profiling studies. It supports large traces generated by tools such as pprof, async-profiler, Java Flight Recorder, and Chrome Tracing exports produced by Chromium and Blink-based browsers. The interface emphasizes quick exploratory analysis comparable to utilities from Apple Inc.'s Instruments and Visual Studio's diagnosis tools, enabling cross-references with commits in systems like GitHub, GitLab, and Bitbucket during performance regressions or benchmarking campaigns.

History and Development

Speedscope grew from demands within communities around V8, Node.js Foundation, and Electron to visualize massive profiles generated by production services at companies including Google, Twitter, Airbnb, and Uber. Early contributions drew on practices established by researchers and engineers from Sun Microsystems, Oracle Corporation, Red Hat, and Intel Corporation working on tracing with SystemTap and oprofile. Development was influenced by visualization approaches pioneered in projects like FlameScope and shaped by discussion in forums such as Stack Overflow and conference proceedings from ACM SIGPLAN, USENIX, QCon, and Velocity. The project accumulated contributions from maintainers affiliated with Mozilla Foundation, Microsoft Research, and independent engineers who previously worked on LLVM, GCC, and JVM performance tools.

Features and Architecture

Speedscope's architecture separates a lightweight web-based visualizer from backend collectors such as perfetto, Linux perf, DTrace, and language-specific profilers like gperftools and async-profiler. The UI renders stacked samples and hierarchical timings using techniques similar to those used in React-based dashboards, D3.js visualizations, and Electron applications. Features include fast searching, differential comparisons reminiscent of git blame workflows in Git, and saving sessions for offline review with file formats compatible with pprof and Chrome Tracing. The codebase incorporates patterns from TypeScript, Webpack, and Babel ecosystems, and its testing and CI pipelines often integrate with services like Travis CI, CircleCI, and GitHub Actions.

File Formats and Compatibility

Speedscope reads multiple profile formats including traces from Chrome Tracing, pprof, Java Flight Recorder, and sampled stack formats emitted by Linux perf and eBPF collectors. It interoperates with exporters and adapters developed in ecosystems such as Go, Java, Python, C++, and Rust. Compatibility layers map per-thread timing data to Speedscope's event model similarly to converters used in FlameGraph toolchains and converters provided by projects like perf-tools and perfcollect. The format flexibility allows integration with observability stacks around OpenTelemetry, Prometheus, and visualization platforms like Grafana when storing or referencing trace artifacts.

Usage and Workflows

Common workflows begin with collecting samples via Linux perf, perfetto, DTrace, or language profilers such as async-profiler for JVM applications and pprof for Go services. Engineers then load traces into Speedscope for hotspot hunting, comparing baseline and regression traces in ways similar to A/B testing of performance patches tracked in GitHub Issues or Jira. Teams integrate trace capture into CI pipelines alongside bazel, Maven, or Gradle builds and use visualization outputs to annotate pull requests in repositories hosted on GitHub, GitLab, or Bitbucket Server. Researchers publishing results at venues like Usenix ATC, ICSE, and PLDI often cite such visual analysis tools when evaluating runtime or compiler optimizations.

Performance and Limitations

Speedscope is optimized for responsive rendering of large traces but relies on upstream collectors (for example perfetto, Linux perf, DTrace) for accurate sampling and minimal overhead. Limitations include dependence on the precision of stack unwinding provided by tools like libunwind and DWARF metadata, challenges with inlined frames in LLVM and GCC-compiled binaries, and reduced visibility in optimized runtimes such as V8 or highly JITted JVM workloads without appropriate debug symbols. Complementary methods using hardware counters from Intel Corporation and AMD processors, or tracing via eBPF on modern Linux kernels, are commonly recommended to mitigate sampling biases and capture wall-clock events more reliably.

Category:Software