Probabilistic SKnife

Probabilistic SKnife (ProbSKnife) is a declarative prototype to evaluate the expected cost of all the partionings of an application with their cost <number_of_domains, future_cost>. To calculate the future cost for all the eligible partitionings, ProbSKnife is launched from a Python script that parses the results, groups them by the labelling, considers the minimum cost to satsify each labelling, and aggregates by eligible partionings.

Prerequisites

Before using ProbSKnife You need to install ProbLog2 and Python.

Tutorial

To try ProbSKnife with our predefined example:

1. Download or clone this repository.

2. Open a terminal in the project folder and run python main.py -h to have the usage of ProbSKnife

   main.py APPID [-d DLIMIT] [-k CHANGELIMIT] [-f] [-l] [-h]"
        APPID           identifier of the application to partition
        DLIMIT          an integer
        CHANGELIMIT     an integer
        -f              shows full results in tables
        -l              shows partitioning labels (impacts on groupby in table)
        -t              shows timestamp of operations
        -h              shows this help
   

3. Execute the program! As an example, running python main.py cloudModel -k 1 -d 5 calculates the costs for the partitionings with at most 5 domains and considering the change of labelling with at most 1 label change of the default example cloudModel.

   The output of such execution is

   Partitioning [[appManager, authenticator, db], [apiGateway], [aiLearning], [userConfig]]                        cost: (4, 3.6789741961)
   Partitioning [[db], [apiGateway], [aiLearning], [appManager, authenticator], [userConfig]]                      cost: (5, 0.0)
   Partitioning [[authenticator, db], [apiGateway], [aiLearning], [appManager], [userConfig]]                      cost: (5, 2.6278387115)
   Partitioning [[appManager, db], [apiGateway], [aiLearning], [authenticator], [userConfig]]                      cost: (5, 6.3068129076)
   Impossible prob: 0.17518924743
   

   Showing the 4 starting partitionings, their cost and the probability to reach a labelling making the application not safely partitionable.

   The probabilistic model with the chosed K is saved in a .pl file to be reused in the future. The format of the file is

   Probability::labelling0L(Labelling).
   

   The above execution creates the file labellingK1_cloudModel.pl with (part) of content

   0.09197435490::labelling0L([(networkData,low),(cryptedData,low),(userPreferences,medium),(userRequests,medium),(iotMeasurements,top),(iotEvents,top),(iotCommands,top),(tlsLibrary,top),(fromProvider,low),(aiFramework,low),(dbms,top),(networkLibrary,low),(dataLibrary,top)]).
   0.09197435490::labelling0L([(networkData,low),(cryptedData,low),(userPreferences,medium),(userRequests,medium),(iotMeasurements,top),(iotEvents,top),(iotCommands,top),(tlsLibrary,top),(fromProvider,low),(aiFramework,low),(dbms,top),(networkLibrary,low),(dataLibrary,low)]).
   

   Showing two labellings with 1 label different from the starting one and their probability.

4. By activating the argument -t the overall results are the same, what changes is the output information view: now the reachable partitionings are annotated with logging information and timestamps.

   As an example, running python main.py cloudModel -k 1 -d 5 -t calculates the costs for the partitionings with at most 5 domains and considering the change of labelling with at most 1 label change of the default example cloudModel.

   The output of such execution is

   Starting execution.
   Retrieving application information.
   Starting to create labelling probabilities at:0:00:00.119381
   
   Labelling created before.
   
   Search for all eligible partitionings.
   4 partitionings determined at: 0:00:00.184154
   
   Determining cost for every partitioning.
   Partitioning [[appManager, authenticator, db], [apiGateway], [aiLearning], [userConfig]]                        cost: (4, 3.6789741961)
   Time from execution start: 0:00:01.465134
   Partitioning [[db], [apiGateway], [aiLearning], [appManager, authenticator], [userConfig]]                      cost: (5, 0.0)
   Time from execution start: 0:00:02.731906
   Partitioning [[authenticator, db], [apiGateway], [aiLearning], [appManager], [userConfig]]                      cost: (5, 2.6278387115)
   Time from execution start: 0:00:04.014362
   Partitioning [[appManager, db], [apiGateway], [aiLearning], [authenticator], [userConfig]]                      cost: (5, 6.3068129076)
   Time from execution start: 0:00:05.279141
   Impossible prob: 0.17518924743
   

   Showing the same output as before with annotated timestamp. Note that the labellaing probabilities where already created to reduce the overall execution time.