While analyzing VSTS load test report we find response times for an endpoint as below
The sampling rate is set to 15 sec as recommended by VSTS (When we set below 15 sec we get a warning). We use VSTS cloud offerings to generate load.
In the performance graph (in results) response time peaks shown are at approx 0.7 seconds and nothing above that. When we download the samples we see all entries(1 entry per 15 seconds) have data matching with what is plotted on the graph and is less that 0.7 seconds.
Our SLA is to support 1 sec for 95%. Now we are unable to understand
For each different counter, one value is generated in each sampling interval and that value is shown on the graphs. Hence the graphs often seem to omit peak values. The tables of results collect all the values and the percentiles for response times are calculated by excluding the slowest values. If a total of N
response times are collected then the 90th percentile excludes the slowest 10% of those N
times, similarly for the other percentiles. To understand the range of response times compare and contrast these values:
Microsoft have several web pages about analysing load test results. This page is one to start with. A web search for terms such as site:microsoft.com analysing load test results
should find much more detail.
The percentile values are calculated by SQL stored procedures that can be found by exploring the load test database. On my Windows 10 computer it is in the files: C:\Users\AdrianHHH\LoadTest2010.mdf
and C:\Users\AdrianHHH\LoadTest2010_log.ldf
. The procedures are also shown in the load test database creation script which should be in C:\Program Files\Microsoft Visual Studio 11.0\Common7\IDE\loadtestresultsrepository.sql
(but change the 11.0
to reflect your Visual Studio version).