InterDomain Path Computation under Domain Uniqueness constraint (IDPCDU) Instances

Thanh Pham Dinh,
Unspecified

Thang Ta Bao,

Hoang Ngo Viet,
Hoang Ngo VietHanoi University of Science & Technology

Long Nguyen Binh
Long Nguyen BinhHanoi University of Science & Technology
Description of this data

The tested data for InterDomain Path Computation under Domain Uniqueness constraint (IDPCDU).

On account of no instances were available for IDPCDU, we made up our minds to generate a set for test instances to evaluate the proposed algorithms. To generate an instance, we first passed three parameter: number of nodes, number of domains and number of edges. After that, we created an array of distinct nodes and an array of distinct domains that satisfied the number of nodes is greater than the length of domain array. Source nodes and terminal node are the first and the last nodes of the nodes array, respectively. With the above arrays, we merged them to make a valid solution called P. Each edge of P was set to the weight one, except the outedge of the source node is set to the weight two. To add noise to the test instance, for every single node in
P, we added some edges to random nodes not in P. Moreover, we created some oneweighededges between the the nodes not in P. These traps would make simple greedy algorithms get it harder to find optimal solution. Eventually, we randomly generated edges that have greater values of weight than the value of the length of P. This method guaranteed that P is the optimal solution of the instance. There were two set of instances created, a small set and a large set. 
Filename idpc_<number of nodes>x<number of domain>x<number of edges>.idpc
First line of a file constains two intergers N and D, which are number of nodes and number of domains, respectively.
Second line contains two integers s and t, which are the source node and terminal node.
Every next line contains four integers u, v, w, d, represents an edge (u,v) has weight w and belong to domain d.
Experiment data files
Steps to reproduce
we made up our minds to generate a set for test instances to evaluate the proposed algorithms. To generate an instance, we first passed three parameter: number of nodes, number of domains and number of edges. After that, we created an array of distinct nodes and an array of distinct domains that satisfied the number of nodes is greater than the length of domain array. Source nodes and terminal node are the first and the last nodes of the nodes array, respectively. With the above arrays, we merged them to make a valid solution called P. Each edge of P was set to the weight one, except the outedge of the source node is set to the weight two. To add noise to the test instance, for every single node in
P, we added some edges to random nodes not in P. Moreover, we created some oneweighededges between the the nodes not in P. These traps would make simple greedy algorithms get it harder to find optimal solution. Eventually, we randomly generated edges that have greater values of weight than the value of the length of P. This method guaranteed that P is the optimal solution of the instance. There were two set of instances created, a small set and a large set.
Related links
Latest version

Version 3
20200118
Published: 20200118
DOI: 10.17632/t726xwcjf9.3
Cite this dataset
Pham Dinh, Thanh; Ta Bao, Thang; Ngo Viet, Hoang; Nguyen Binh, Long (2020), “InterDomain Path Computation under Domain Uniqueness constraint (IDPCDU) Instances”, Mendeley Data, v3 http://dx.doi.org/10.17632/t726xwcjf9.3
Statistics
Previous versions
Compare to version
Institutions
Categories
Licence
The files associated with this dataset are licensed under a Creative Commons Attribution 4.0 International licence.