Indeed. Often latency measurements are clumpy. You might have one clump of fast responses when no GC occurs, another when some GC occurs, and a smaller clump

Indeed. Often latency measurements are clumpy. You might have one clump of fast responses when no GC occurs, another when some GC occurs, and a smaller clump where a stop-the-world full GC has occurred. In such a case average is not meaningful.