Boost application performance with profile guided optimization Michael Pratt, Google

Profile Guided Optimization (PGO) is a compiler optimization technique that uses runtime behavior data to improve application performance, typically achieving 2-15% CPU improvements

Key optimizations PGO enables:

• Function inlining decisions
• Escape analysis for better stack vs heap allocation
• Constant propagation and folding
• Focus optimization efforts on hot code paths

Best practices for using PGO:

• Collect profiles from real production environment rather than benchmarks
• Keep profiles alongside source code in repository
• Regularly update profiles as application evolves
• Profile for at least 30 seconds to get representative data

PGO is designed to be:

• Source-stable (works with small code changes)
• Iteratively stable (can profile PGO-optimized binaries)
• Safe against profile mismatches (won’t degrade performance)
• Easy to use with minimal configuration

Implementation details:

• Uses pprof for profile collection and analysis
• Profiles should be named default.pgo in package directory
• Works automatically with Go build once profile is in place
• Can be enabled via -pgo flag for explicit profile path

Most impactful optimizations come from:

• Moving heap allocations to stack
• Optimizing hot functions identified in profile
• Improving performance of dependencies and standard library code
• Focusing on CPU-bound workloads

Practical considerations:

• Typically reduces container CPU time and associated costs
• Binary size may increase slightly from inlining
• Works best with continuous profiling in production
• Effects vary by application (2-15% improvement range)