Supercharging Redpanda Streaming with profile-guided optimization

In the release of Redpanda Streaming 26.1, the team explored optimizing the binary using profile-guided optimization (PGO) and LLVM BOLT to enhance performance. PGO and BOLT are technologies that refine application binaries based on profiling data, but differ in their methods; PGO involves a two-phase compilation with instrumentation, while BOLT optimizes post-link binaries directly. Despite BOLT's potential for saving build time, its brittleness led to bugs, prompting the decision to focus on the well-established PGO, which has been shown to significantly enhance CPU-bound performance by reducing latencies and improving CPU utilization. The use of PGO resulted in up to 47% lower latencies and 15% better CPU reactor utilization in core regression benchmarks. A top-down performance analysis (TMA) highlighted that PGO effectively addresses frontend-bound CPU issues by improving code locality, thus enhancing instruction cache efficiency and execution rates. The benefits of PGO, as demonstrated through heatmap visualizations of code access frequency, are integral to Redpanda Streaming's improved performance in version 26.1, making it a recommended optimization for CPU-intensive workloads.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.