From 0f6410d3c95eb28164fdd349778155c6525e1a79 Mon Sep 17 00:00:00 2001 From: Tibor Frank Date: Mon, 22 Jan 2018 15:43:22 +0100 Subject: CSIT-902: Compare performance results before/after meltdown/spectre - CSIT-903: LLD - CSIT-904: Data model - CSIT-905: Algorithm - CSIT-906: Static content Change-Id: Ia7b77fc35ab852110c2f50efb7756ac15576749a Signed-off-by: Tibor Frank Signed-off-by: Maciek Konstantynowicz --- resources/tools/presentation/doc/pal_lld.rst | 116 ++++++++++++++++++++++----- 1 file changed, 98 insertions(+), 18 deletions(-) (limited to 'resources/tools/presentation/doc') diff --git a/resources/tools/presentation/doc/pal_lld.rst b/resources/tools/presentation/doc/pal_lld.rst index 027d6b326a..64bde3e5fe 100644 --- a/resources/tools/presentation/doc/pal_lld.rst +++ b/resources/tools/presentation/doc/pal_lld.rst @@ -1109,8 +1109,9 @@ For example, the element which specification includes: filter: - "'64B' and 'BASE' and 'NDRDISC' and '1T1C' and ('L2BDMACSTAT' or 'L2BDMACLRN' or 'L2XCFWD') and not 'VHOST'" -will be constructed using data from the job "csit-vpp-perf-1707-all", for all listed -builds and the tests with the list of tags matching the filter conditions. +will be constructed using data from the job "csit-vpp-perf-1707-all", for all +listed builds and the tests with the list of tags matching the filter +conditions. The output data structure for filtered test data is: @@ -1189,6 +1190,83 @@ Subset of existing performance tests is covered by TSA graphs. "plot-throughput-speedup-analysis" +Comparison of results from two sets of the same test executions +''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''' + +This algorithm enables comparison of results coming from two sets of the +same test executions. It is used to quantify performance changes across +all tests after test environment changes e.g. Operating System +upgrades/patches, Hardware changes. + +It is assumed that each set of test executions includes multiple runs +of the same tests, 10 or more, to verify test results repeatibility and +to yield statistically meaningful results data. + +Comparison results are presented in a table with a specified number of +the best and the worst relative changes between the two sets. Following table +columns are defined: + + - name of the test; + - throughput mean values of the reference set; + - throughput standard deviation of the reference set; + - throughput mean values of the set to compare; + - throughput standard deviation of the set to compare; + - relative change of the mean values. + +**The model** + +The model specifies: + + - type: "table" - means this section defines a table. + - title: Title of the table. + - algorithm: Algorithm which is used to generate the table. The other + parameters in this section must provide all information needed by the used + algorithm. + - output-file-ext: Extension of the output file. + - output-file: File which the table will be written to. + - reference - the builds which are used as the reference for comparison. + - compare - the builds which are compared to the reference. + - data: Specify the sources, jobs and builds, providing data for generating + the table. + - filter: Filter based on tags applied on the input data, if "template" is + used, filtering is based on the template. + - parameters: Only these parameters will be put to the output data + structure. + - nr-of-tests-shown: Number of the best and the worst tests presented in the + table. Use 0 (zero) to present all tests. + +*Example:* + +:: + + - + type: "table" + title: "Performance comparison" + algorithm: "table_performance_comparison" + output-file-ext: ".csv" + output-file: "{DIR[DTR,PERF,VPP,IMPRV]}/vpp_performance_comparison" + reference: + title: "csit-vpp-perf-1801-all - 1" + data: + csit-vpp-perf-1801-all: + - 1 + - 2 + compare: + title: "csit-vpp-perf-1801-all - 2" + data: + csit-vpp-perf-1801-all: + - 1 + - 2 + data: + "vpp-perf-comparison" + filter: "all" + parameters: + - "name" + - "parent" + - "throughput" + nr-of-tests-shown: 20 + + Advanced data analytics ``````````````````````` @@ -1216,7 +1294,8 @@ Tables - tables are generated by algorithms implemented in PAL, the model includes the algorithm and all necessary information. - output format: csv - - generated tables are stored in specified directories and linked to .rst files. + - generated tables are stored in specified directories and linked to .rst + files. Plots @@ -1232,8 +1311,8 @@ Report generation ----------------- Report is generated using Sphinx and Read_the_Docs template. PAL generates html -and pdf formats. It is possible to define the content of the report by specifying -the version (TODO: define the names and content of versions). +and pdf formats. It is possible to define the content of the report by +specifying the version (TODO: define the names and content of versions). The process @@ -1251,12 +1330,13 @@ The process 5. Generate the report. 6. Store the report (Nexus). -The process is model driven. The elements’ models (tables, plots, files and -report itself) are defined in the specification file. Script reads the elements’ -models from specification file and generates the elements. +The process is model driven. The elements' models (tables, plots, files +and report itself) are defined in the specification file. Script reads +the elements' models from specification file and generates the elements. -It is easy to add elements to be generated, if a new kind of element is -required, only a new algorithm is implemented and integrated. +It is easy to add elements to be generated in the report. If a new type +of an element is required, only a new algorithm needs to be implemented +and integrated. API @@ -1396,12 +1476,12 @@ PAL functional diagram How to add an element ````````````````````` -Element can be added by adding its model to the specification file. If the -element will be generated by an existing algorithm, only its parameters must be -set. +Element can be added by adding it's model to the specification file. If +the element is to be generated by an existing algorithm, only it's +parameters must be set. -If a brand new type of element will be added, also the algorithm must be -implemented. -The algorithms are implemented in the files which names start with "generator". -The name of the function implementing the algorithm and the name of algorithm in -the specification file had to be the same. +If a brand new type of element needs to be added, also the algorithm +must be implemented. Element generation algorithms are implemented in +the files with names starting with "generator" prefix. The name of the +function implementing the algorithm and the name of algorithm in the +specification file have to be the same. -- cgit 1.2.3-korg