DependencyTransformation

• Dependency transformation computes two way dependency in non oriented weighted graph.
• Input for this application is list of edges with their weight, in format IdA;IdB;Weight
• Result of Dependency transformation is non symetric matrix of dependencies.

Parallel computation

In this program we use multiple method how to optimize performance of algorithm and compare their performance.

1. Sequentional approach
2. Native Microsoft System.Threading.TasksParallel.For
3. Microsoft Task
4. Implementation of Parallel manager - Parallel-For
5. Naive parallelism

Tests

We use XUnit to ensure correct results of Dependecy calculation. There is several Test Cases that tests:

• Loading graph from csv
• Computation of dependency for one node (cell of matrice)
• Computation of dependency on small sample matrix
• Computation of dependency on real data with comparison to precomputed results

Results

Table of average ticks per computation.

	Lesmis	Karate	USAirports	Newman	as-22july06	astro-ph	Decagon
Sequential	6960	6991	178667	91008613	12692952	17202213	376663664
Microsofrt Task	5635	6057	116831	54133513	6859156	7776491	94403879
Parallel Processor (own impl.)	3870	6057	528804	66788621	4470979	1655129	28239280
Microsoft Parallel.For	9437	5576	155050	66153578	16643949	10837409	142772174
Naive Paralell Threads	210272	90015	2564208	NAN	NAN	NAN	NAN

* https://toreopsahl.com/datasets/#usairports

** https://toreopsahl.com/datasets/#newman2001

Note: Number is average per 5 runs of alg.