M3, an open-source distributed metrics engine, is relied upon by major companies as their mission critical solution to monitoring their software at a global scale. As metrics volume continues to grow into the order of billions of data points being read/written per second, open source contributors are exploring and implementing new approaches (some novel, and some simple) to minimize memory and improve query/ingestion speeds. These improvements allow M3 users to push further the boundary of metrics volume (~10s of billions of metrics) for even less compute cost. In the talk, we’ll dive into a few of these performance optimizations, focusing on how contributors went about identifying bottlenecks, exploring potential solutions, benchmarking results, and testing for regressions.