Add aligned system comparison aggregated charts

[spi-x-i]

Main Goal

Provide a new API able to:

• select multiple source directories (i.e. from different systems)
• produce paper-shaped aggregation charts moved to an aligned context, i.e. keeping x and y axis aligned to same values

  e.g. assume system A runtime is 30s and system B runtime is 45s; conversely system B focused metric - assume disk - goes up to 30 MB - against 200 MB upper bound of system B, we want achieve x aligned to 45s and y to 200 MB

• save these additional charts to directory aligned

Current State

Two aggregation features:

• aggregated paper-shaped charts: cpu, disk, memory, network
• aggregated by metric and sub metric based on average and standard deviation measures

Draft Design

• sources path must be from different systems, hence include a check here; it could be: assure these are not identical paths
• load data
• build a function align which takes as input two df and returns the x/y upper bounds
• produces aggregated object for both systems
• call plot functions passing as input the bounds
• save these new charts in an aligned directory within the system parent directory; the charts filenames should be suffixed by a datetime

[spi-x-i]

Update

Currently the feature is temporary implemented by branch aggregateB.

How to Use

-cmp option allows to produce aligned results between provided data sources; the -cmp option works on top of aggregation solution only. The API considers the target directory (i.e. the current dir) as container of experiments to compare.

Assume you have a dir A containing dir X and Y where X and Y are two different experiments dirs, shee will fetch data over 'logs/dstat/dstat-0.7.2' for both X and Y and will produce two output dir for both X and Y:

• aggregation > the usual aggregation charts
• _aggregationbenchmark > the aligned version

Call the API as follows:

python -m shee -O -a -cmp

Thanks to @FelixNeutatz contribute.

[spi-x-i]

Anyway the method should be reimplemented in order to:

• connect the concurrent experiments charts
• untick the x-axis of the experiment with shortest runtime after the runtime end in order to keep time range clearer
• separate methods and some refactoring