Skip to content

Latest commit

 

History

History
181 lines (148 loc) · 5.07 KB

README.md

File metadata and controls

181 lines (148 loc) · 5.07 KB
Raw

banner

Resources utilisation plot

Visualise the resources used by different modules in a CMSSW job as a hierarchical chart.

This tool is composed of two parts:

  1. running a CMSSW job and producing a FastTimerService output in JSON format;
  2. uploading the data to a web server for interactive visualisation and exploration.

Running a CMSSW job

The instructions for running a CMSSW job with the FastTimerService are available on its twiki page.

To produce a resources.json file suitable for uploading and visualisation, use CMSSW_11_1_0_pre5 or later with the options

process.FastTimerService.writeJSONSummary = cms.untracked.bool(True)
process.FastTimerService.jsonFileName = cms.untracked.string('resources.json')

The resources.json file should look like

{
  "modules": [
    {
      "events": 100,
      "label": "source",
      "mem_alloc": 302244,
      "mem_free": 151125,
      "time_real": 120.584236,
      "time_thread": 120.307255,
      "type": "FedRawDataInputSource"
    },
    ...
  ],
  "resources": [
    {
      "time_real": "real time"
    },
    {
      "time_thread": "cpu time"
    },
    {
      "mem_alloc": "allocated memory"
    },
    {
      "mem_free": "deallocated memory"
    }
  ],
  "total": {
    "events": 100,
    "label": "HLTGRun",
    "mem_alloc": 27584018,
    "mem_free": 27066780,
    "time_real": 59625.846558,
    "time_thread": 59310.740262,
    "type": "Job"
  }
}

The file is structured in three sections:

  • resources lists the available metrics and their brief descriptions;
  • modules is an array with the type, label and metrics for each module in the process;
  • total lists the information and metrics for the whole job.

Setting up the interactive page on a web server

After cloning the repository, the directory structure should look like this:

scripts/
    convert.py
    merge.py
web/
    Buttons-1.5.6/
    CarrotSearch/
    DataTables-1.10.18/
    FixedColumns-3.2.5/
    FixedHeader-3.1.4/
    RowGroup-1.1.0/
    Scroller-2.0.0/
    cgi-bin/
        colourslist.py
        datalist.py
        groupslist.py
    colours/
        default.json
    data/
    groups/
        hlt_cpu.json
        hlt_gpu.json
        reco_PhaseII.json
    jQuery-3.3.1/
    common.js
    datatables.css
    datatables.js
    datatables.min.css
    datatables.min.js
    piechart.html
    piechart.html
README.md

Just copy the web/ drectory to a web server. Two versions of the web page are available:

  • piechart.php requires PHP to be enabled in the web server;
  • piechart.html does not use PHP, but requires support for cgi-bin scripts for the cgi-bin directory.

The functionality of the two pages is otherwise identical.

Visualising the data

To make a measurement available on the web server, copy the JSON files produced by the FastTimerService to the data/ subdirectory, and refresh the web page. The files will automatically appear in the "Datasets" drop-down box.

To look at a measurement without adding it permanently to the web server, use the "upload a file" button: this will upload and visualise the dataset, without adding it permanently to the web server.

The metric, grouping and colour style can be changed directly on the web page.

Working with JSON files

While the JSON files produces by the FastTimerService can be visualised directly in the web interface, under the scripts/ directory there some python scripts to perfrm common operations on them: merging multiple files, converting from the old format, etc.

Merging multiple JSON files

It is possible to merge multiple JSON files, for exmple to "harvest" the results of multiple jobs running in paralle, using the merge.py script.

./scripts/merge.py input1.json input2.json ... > output.json

To be mergeable, the files must list the same metrics (i.e. the resources section of the JSON).

Converting old JSON files

The format of the JSON files used by the web interface has changed with respect to what was produced by make_circles.py.

While the old files contained less information, it is possible to use the script convert.py (found in the scripts/ subdirectory) to convert them to the latest format and visualise them on the web interface, and e.g. change the grouping and colour scheme on fly:

./scripts/convert.py old.json > new.json

Colouring a dependency graph

The script dot_colour.py can be used to apply the same groups and colour scheme used by a pie chart to a CMSSW dependency graph, generated by the DependencyGraph service. The groups and colour scheme to use can be specified via the -g and -c options:

./scripts/dot_colour.py -g hlt -c default dependency.dot > coloured.dot

Finding unassigned entries

The script find_unassigned.py can be used to find all entries module type|module label in a JSON file that are not assigned to any group:

./scripts/find_unassigned.py resurces.json