PerfCells: A case study in achieving clean performance test results By Vivek Venkatachalam ________________________________________________________________________________________________________________ Copies may not be made or distributed for commercial use Excerpt from PNSQC 2011 PNSQC.ORG 1
Performance is key for quality. ________________________________________________________________________________________________________________ Copies may not be made or distributed for commercial use Excerpt from PNSQC 2011 PNSQC.ORG 2
Inconsistent results makes it hard to do performance testing well. ________________________________________________________________________________________________________________ Copies may not be made or distributed for commercial use Excerpt from PNSQC 2011 PNSQC.ORG 3
The real problems are hard to find ________________________________________________________________________________________________________________ Copies may not be made or distributed for commercial use Excerpt from PNSQC 2011 PNSQC.ORG 4
What if you could eliminate the haystack ? ________________________________________________________________________________________________________________ Copies may not be made or distributed for commercial use Excerpt from PNSQC 2011 PNSQC.ORG 5
First, an overview of our performance engineering system - Planning - Instrumentation - Testing - Analysis ________________________________________________________________________________________________________________ Copies may not be made or distributed for commercial use Excerpt from PNSQC 2011 PNSQC.ORG 6
Planning - Key customer scenarios ________________________________________________________________________________________________________________ Copies may not be made or distributed for commercial use Excerpt from PNSQC 2011 PNSQC.ORG 7
Planning - Metrics ________________________________________________________________________________________________________________ Copies may not be made or distributed for commercial use Excerpt from PNSQC 2011 PNSQC.ORG 8
Instrumentation – Inserting code markers ________________________________________________________________________________________________________________ Copies may not be made or distributed for commercial use Excerpt from PNSQC 2011 PNSQC.ORG 9
Testing - Manual ________________________________________________________________________________________________________________ Copies may not be made or distributed for commercial use Excerpt from PNSQC 2011 PNSQC.ORG 10
Testing - Automation ________________________________________________________________________________________________________________ Copies may not be made or distributed for commercial use Excerpt from PNSQC 2011 PNSQC.ORG 11
Analysis – Processing the results ________________________________________________________________________________________________________________ Copies may not be made or distributed for commercial use Excerpt from PNSQC 2011 PNSQC.ORG 12
The problem: Lots of noise in the system. ________________________________________________________________________________________________________________ Copies may not be made or distributed for commercial use Excerpt from PNSQC 2011 PNSQC.ORG 13
Unexplainable variations in results led to wasted time and effort ________________________________________________________________________________________________________________ Copies may not be made or distributed for commercial use Excerpt from PNSQC 2011 PNSQC.ORG 14
Needed a quick fix ________________________________________________________________________________________________________________ Copies may not be made or distributed for commercial use Excerpt from PNSQC 2011 PNSQC.ORG 15
Decided to focus only on large anomalies ________________________________________________________________________________________________________________ Copies may not be made or distributed for commercial use Excerpt from PNSQC 2011 PNSQC.ORG 16
The long term solution ________________________________________________________________________________________________________________ Copies may not be made or distributed for commercial use Excerpt from PNSQC 2011 PNSQC.ORG 17
Brainstorming ________________________________________________________________________________________________________________ Copies may not be made or distributed for commercial use Excerpt from PNSQC 2011 PNSQC.ORG 18
Realization : We had conflicting goals Prevent Verify that regressions real world however performance small. goals were being met. ________________________________________________________________________________________________________________ Copies may not be made or distributed for commercial use Excerpt from PNSQC 2011 PNSQC.ORG 19
The solution: Perf cells ________________________________________________________________________________________________________________ Copies may not be made or distributed for commercial use Excerpt from PNSQC 2011 PNSQC.ORG 20
Key point: one machine ________________________________________________________________________________________________________________ Copies may not be made or distributed for commercial use Excerpt from PNSQC 2011 PNSQC.ORG 21
Advantages - Isolated ________________________________________________________________________________________________________________ Copies may not be made or distributed for commercial use Excerpt from PNSQC 2011 PNSQC.ORG 22
Advantages - Reliable behavior ________________________________________________________________________________________________________________ Copies may not be made or distributed for commercial use Excerpt from PNSQC 2011 PNSQC.ORG 23
Advantages - Easy management ________________________________________________________________________________________________________________ Copies may not be made or distributed for commercial use Excerpt from PNSQC 2011 PNSQC.ORG 24
Advantages - Scale Out ________________________________________________________________________________________________________________ Copies may not be made or distributed for commercial use Excerpt from PNSQC 2011 PNSQC.ORG 25
Our results ________________________________________________________________________________________________________________ Copies may not be made or distributed for commercial use Excerpt from PNSQC 2011 PNSQC.ORG 26
Histogram of elapsed times from tests on pre-production topology ________________________________________________________________________________________________________________ Copies may not be made or distributed for commercial use Excerpt from PNSQC 2011 PNSQC.ORG 27
Histogram of elapsed times from tests on a PerfCell ________________________________________________________________________________________________________________ Copies may not be made or distributed for commercial use Excerpt from PNSQC 2011 PNSQC.ORG 28
Key Takeaways • Steady losses add up. • PerfCells increased test sensitivity. • Allowed us to accept variability in real world testing. ________________________________________________________________________________________________________________ Copies may not be made or distributed for commercial use Excerpt from PNSQC 2011 PNSQC.ORG 29
Some possible improvements • Test builds simultaneously • Partition tests • PerfCells in the cloud ________________________________________________________________________________________________________________ Copies may not be made or distributed for commercial use Excerpt from PNSQC 2011 PNSQC.ORG 30
Credits My co-authors Marcelo and Shirley. Our reviewers Bob and Mano Microsoft Clip Art for the images. ________________________________________________________________________________________________________________ Copies may not be made or distributed for commercial use Excerpt from PNSQC 2011 PNSQC.ORG 31
Recommend
More recommend