Richard Startin`, a Senior Software Engineer at `Datadog`, shares his experience on using the Java wallclock profiler to improve request latency in Java applications. He highlights the limitations of traditional profiling approaches and introduces `async-profiler`, which he worked on to help address this issue. The article explains how `async-profiler` can be used to create a flame graph that shows a complete picture of CPU time, even if the bottleneck is not directly related to CPU usage. It also discusses the challenges of sampling threads in a high-thread-count environment and introduces a new thread filter approach that only samples threads with an active trace context set up by the tracer. The article concludes with examples of how to use `async-profiler` to investigate and fix performance issues, including increasing the heap size and adding more vCPUs.