Calculate Histogram diffs rather reporting all to Traces
We currently measure Histograms in perf benchmarks. This is done by emitting a trace of all histograms from startup. Then emitting another at the end of tracing. Then in Catapult we calculate differences in uma_metric.html However this has led to issues in the past with mis-calculating buckets of histograms. It also currently breaks down when histogram emitting changes occur near the startup. To help clear this, we want to not record the initial histograms. But instead calculate the differences. Thus emitting only one sample for Catapult to process. We will then reduce the complexity of uma_metric's processing in a follow up change in Catapult. Bug: 1145731 Change-Id: Ia6452984a16914aaf19e66c22d583592835f46df Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/2521106Reviewed-by:oysteine <oysteine@chromium.org> Commit-Queue: Jonathan Ross <jonross@chromium.org> Cr-Commit-Position: refs/heads/master@{#827962}
Showing
Please register or sign in to comment