diff options
author | jasplin <qt-info@nokia.com> | 2011-01-19 11:31:46 +0100 |
---|---|---|
committer | jasplin <qt-info@nokia.com> | 2011-01-19 11:31:46 +0100 |
commit | 2582b5ad4c3360ec63eb82df4253c93a0de19757 (patch) | |
tree | 5e604d78ab96974df9407d0543ef293ee57f6ee1 /web/getstats/help_tsbm.html | |
parent | 7af7ec23e722e315420c36268a85b45a2de63f13 (diff) |
Updated online help documentation.
Diffstat (limited to 'web/getstats/help_tsbm.html')
-rw-r--r-- | web/getstats/help_tsbm.html | 365 |
1 files changed, 365 insertions, 0 deletions
diff --git a/web/getstats/help_tsbm.html b/web/getstats/help_tsbm.html new file mode 100644 index 0000000..81fd50d --- /dev/null +++ b/web/getstats/help_tsbm.html @@ -0,0 +1,365 @@ +<html> + +<head> + +<style type="text/css"> +table { border-collapse:collapse; } +table, th, td { border: 1px solid #aaaaaa; } +th {vertical-align:top; padding:10px} +td {vertical-align:top; padding:10px} +li {padding:5px} +</style> + +</head> + +<body> + +<span style="text-align:center"> +<h1>BM2 Documentation</h1> +<h2><em>Benchmark Time Series Page</em></h2> +</span> + + +<!-- ///////////////////////////////////////////////////////////////// --> +<h2>Overview</h2> + <img src="images/help_overview_anno.png" /><br /><br /> + +<h3>Main context</h3> + +<b>Note:</b> Some of the terms and concepts in the below table are described +in greater detail elsewhere. The additional documentation can often be +accessed through a link. +<br /><br /> + +<table> + <tr> + <td style="text-align:right"><b>Database</b></td> + <td> + The database from which results for the time series were extracted. + </td> + </tr> + <tr> + <td style="text-align:right"><b>Report date</b></td> + <td> + The date at which the web page was generated. + </td> + </tr> + <tr> + <td style="text-align:right"><b>Host</b></td> + <td> + The physical computer on which the benchmark producing this time series + was executed. <b>Note:</b>It is assumed that the HW/SW specifications of + the host does not change significantly during the time span of the time + series. (The principle here being that significant changes in the time + series should be caused by changes in the product (i.e. Qt) only!) + </td> + </tr> + <tr> + <td style="text-align:right"><b>Platform</b></td> + <td> + The general environment used for building and executing the product + being measured (typically an OS/compiler combination). + </td> + </tr> + <tr> + <td style="text-align:right"><b>Branch</b></td> + <td> + Essentially the version of the product being measured. The branch is + normally made up of two components: The <i>git repository</i> and the + <i>git branch</i>. + </td> + </tr> + <tr> + <td style="text-align:right"><b>Target snapshots</b></td> + <td> + The requested subsequence of snapshots for which results for this + host/platform/branch/benchmark/metric combination potentially exist in + the database. + </td> + </tr> + <tr> + <td style="text-align:right"><b>Difference tolerance</b></td> + <td> + A real value ≥ 1 that decides whether a <a href="#changes">change</a> + between two median observations in the time series is considered + significant or not. + </td> + </tr> + <tr> + <td style="text-align:right"><b>Minimum durability tolerance</b></td> + <td> + The minimum length a contiguous sequence of significantly equal median + observations must have for it to achieve a durability score greater + than zero. Once the sequence is at least this long, the durability + score grows linearly to 1 at a rate that depends on the maximum + durability tolerance. + </td> + </tr> + <tr> + <td style="text-align:right"><b>Maximum durability tolerance</b></td> + <td> + The length of a contiguous sequence of significantly equal median + observations that is sufficient to achieve the maximum durability score + of 1. The durability score for shorter sequences falls linearly to 0 at + a rate that depends on the minimum durability tolerance. + </td> + </tr> +</table> + + +<h3>Time series statistics</h3> + +<b>Note:</b> Some of the terms and concepts in the below table are described +in greater detail elsewhere. The additional documentation can often be +accessed through a link. +<br /><br /> + +<table> + <tr> + <td style="text-align:right"><b>MS</b></td> + <td> + Missing snapshots, i.e. the number of target snapshots for which no + results exist. + <br /><br />A high value might indicate unstable execution. + </td> + </tr> + <tr> + <td style="text-align:right"><b>LSD</b></td> + <td> + Last Snapshot Distance, i.e. distance between the last target snapshot + and the last snapshot in the time series. + <br /><br />If the last target snapshot is the last one available in the + database, a high value might indicate that the benchmark currently fails + to produce results. + </td> + </tr> + <tr> + <td style="text-align:right"><b>NI</b></td> + <td> + Total number of observations explicitly flagged as invalid. + <br /><br />An invalid observation is typically caused by a failed + QVERIFY() etc. + </td> + </tr> + <tr> + <td style="text-align:right"><b>NZ</b></td> + <td> + Total number of non-positive observations. + <br /><br />Normally an observation must be positive to be valid. + <br /><b>Note:</b> A non-positive observation is not necessarily flagged + as <i>invalid</i> (see <b>NI</b>). + </td> + </tr> + <tr> + <td style="text-align:right"><b>NC</b></td> + <td> + Number of <a href="#changes">significant changes</a>. + <br /><br />A high value might indicate unstable or fluctuating results. + </td> + </tr> + <tr> + <td style="text-align:right"><b>MDRSE</b></td> + <td> + Median of the + valid <a href="http://en.wikipedia.org/wiki/Standard_error_%28statistics%29#Relative_standard_error">relative + standard errors</a> of all snapshots. + <br /><br /> + + <img src="images/rse.png" /> + <br /><br />A high value might indicate unstable or fluctuating results. + </td> + </tr> + <tr> + <td style="text-align:right"><b>RSEMD</b></td> + <td> + Relative standard error (see above) of the valid median observations of + all snapshots. + <br /><br />A high value might indicate either 1) unstable or + fluctuating results or 2) stable changes of a high magnitude. + </td> + </tr> + <tr> + <td style="text-align:right"><b>LC</b></td> + <td> + Last <a href="#changes">significant change</a>. + <br /><br />The higher the value is above 1, the more strongly it + represents an improvement. + <br />The lower the value is below 1, the more strongly it represents a + regression. + </td> + </tr> + <tr> + <td style="text-align:right"><b>LCDA</b></td> + <td> + Days ago (relative to the report date) since the first observation for + the last <a href="#changes">significant change</a> snapshot was uploaded + to the database. + <br />The distance (in terms of number of target snapshots) between the + last significant change snapshot and the last target snapshot is shown + in parentheses. + </td> + </tr> + <tr> + <td style="text-align:right"><b>LCMS</b></td> + <td> + Magnitude score of the last <a href="#changes">significant + change</a>. This score indicates the strength of the last signicifant + change as a value ranging from 0 (weak) to 1 (strong): + <br /><br /> + <img src="images/lcms.png" /> + </td> + </tr> + <tr> + <td style="text-align:right"><b>LCGSS</b></td> + <td> + Global separation score for the last <a href="#changes">significant + change</a>. This score indicates how well the median observation at the + last significant change snapshot are separated from the median + observations at <u>all preceding</u> snapshots in the time series. The + median observation at the last <a href="#changes">base snapshot</a> is + used as the maximum separation reference. + + <br /><br />The score ranges from 0 (weak separation) to 1 (strong + separation). + + <br /><br />This score roughly measures how close the median observation + at the last significant change is to represent an "all time high(low)" + up to this point in the history. + </td> + </tr> + <tr> + <td style="text-align:right"><b>LCLSS</b></td> + <td> + Local separation score for the last <a href="#changes">significant + change</a>. This score indicates how well the median observations on + each side of the last significant change snapshot are separated from + each other. Snapshots before the last <a href="#changes">base + snapshot</a> are not considered. The median observation at the base + snapshot is used as the maximum separation reference. + + <br /><br />The score ranges from 0 (weak separation) to 1 (strong + separation). + </td> + </tr> + <tr> + <td style="text-align:right"><b>LCDS1</b></td> + <td> + Durability score 1 for the last <a href="#changes">significant + change</a>. This score indicates the distance (in terms of number of + snapshots) from the last significant change to + its <a href="#changes">base snapshot</a>. + + <br /><br />The score ranges from 0 (weak durability) to 1 (strong + durability) and is scaled against the min/max durability tolerances. + + <br /><br />This score measures for how long the median observation + stayed near the base value until the last significant change occurred. + </td> + </tr> + <tr> + <td style="text-align:right"><b>LCDS2</b></td> + <td> + Durability score 2 for the last <a href="#changes">significant + change</a>. This score indicates the distance (in terms of number of + snapshots) from the last significant change to the end of the time + series. + + <br /><br />The score ranges from 0 (weak durability) to 1 (strong + durability) and is scaled against the min/max durability tolerances. + + <br /><br />This score measures for how long the median observation at + the last significant change has stayed essentially the same. + </td> + </tr> + <tr> + <td style="text-align:right"><b>LCSS</b></td> + <td> + Stability score for the last <a href="#changes">significant change</a>: + <br /><br /> + + <b>LCMS</b> * <b>LCGSS</b> * <b>LCLSS</b> * <b>LCDS1</b> * <b>LCDS2</b> + <br /><br /> + The higher this score, the higher the likelihood that the last + significant change is or will become permanent. + </td> + </tr> + <tr> + <td style="text-align:right"><b>LCSS1</b></td> + <td> + Stability score for the last <a href="#changes">significant change</a> + that does not consider the history after the latter: + <br /><br /> + + <b>LCMS</b> * <b>LCGSS</b> * <b>LCLSS</b> * <b>LCDS1</b> + <br /><br /> + The higher this score, the higher the likelihood that the last + signicifant change is or will become permanent, but since <b>LCDS2</b> + is omitted from the product, a high <b>LCSS1</b> is more likely to be + caused by an outlier than a high <b>LCSS</b>! + </td> + </tr> +</table> + + + +<h3>Benchmark and metric</h3> +The benchmark name consists of three subnames and is formatted like this: +<br /><br /> +<subname 1><b>:</b><subname 2><b>(</b><subname 3><b>)</b> +<br /><br /> + +The subnames can essentially be anything not containing the characters +'<b>:</b>', '<b>(</b>', and '<b>)</b>'. Only subname 3 may contain whitespace. + +For benchmark results generated by QTestLib, the subnames always correspond to +<i>test case</i>, <i>test function</i>, and <i>data tag</i> respectively. + +<br /><br /> +The metric name is one of a set of predefined metric names, each of which is +classified as either "lower is better" (like walltime) or "higher is better" +(like fps). + +<!-- ///////////////////////////////////////////////////////////////// --> +<h2>Time Series Plot</h2> + +<h3>Snapshots and main graph</h3> + <img src="images/help_plot_overview_anno.png" /><br /><br /> + + +<a name="changes" /> +<h3>Significant changes</h3> + <img src="images/help_plot_changes_anno.png" /><br /><br /> + + +<h3>Sample size</h3> + <img src="images/help_plot_samplesize_anno.png" /><br /><br /> + + +<h3>Non-positive observations</h3> + <img src="images/help_plot_nonposobs_anno.png" /><br /><br /> + + +<h3>Invalid observations</h3> + <img src="images/help_plot_invalidobs_anno.png" /><br /><br /> + + +<h3>Statistical dispersion</h3> +Statistical dispersion in a time series is measured in terms of +<a href="http://en.wikipedia.org/wiki/Standard_error_%28statistics%29#Relative_standard_error">relative standard error</a> (RSE). + <img src="images/help_plot_rse_anno.png" /><br /><br /> + + +<h3>Missing data</h3> + <img src="images/help_plot_missing_anno.png" /><br /><br /> + + +<h2>Snapshot details</h2> +When clicking on a snapshot in the plot, the two tables below the plot are +filled with various details about the selected snapshot. +<br /> +<br /> + <img src="images/help_plot_selected_anno.png" /><br /><br /> + +</body> + +</html> |