model ModelName
options noimplicit
uses "mmxprs" ,"mmsystem"


  declarations
    sites = 1..15 
    facilities = 1..10
    facilities_possible=5  
    start_time, end_time: real
    DISTANCES: array(sites,facilities) of real
    Y:  array(sites,facilities) of mpvar
    X: array(facilities) of mpvar
    Z:mpvar
    Objective:linctr
  end-declarations

  initialisations from "input.dat"
    DISTANCES
  end-initialisations

  forall(i in sites,j in facilities) Y(i,j) is_binary

  forall(j in facilities) X(j) is_binary

  (sum(j in facilities) X(j))=facilities_possible

  forall(i in sites) (sum(j in facilities) Y(i,j))=1

  forall(i in sites,j in facilities) Y(i,j)<=X(j)

  forall(j in facilities) (sum(i in sites) Y(i,j) * DISTANCES(i,j))<=Z


  Objective:=Z
  !exportprob("model.lp")
  !setparam("XPRS_VERBOSE", true)



  start_time := gettime
  minimise(Objective)
  end_time := gettime
  writeln("nTime: ", (end_time - start_time), " s")

  writeln("facilities:")
  forall(i in facilities|getsol(X(i))=1) do
    writeln(i)  
  end-do

  writeln("y")
  forall(i in sites, j in facilities|getsol(Y(i, j))=1) do
      writeln("For site " + i + " sito " + j + " assigned: " + getsol(Y(i, j)))
  end-do

  writeln(" " + getsol(Objective))


end-model

I wrote this problem, my goal is to find the difficult instances of it , then the input matrices (DISTANCES) that take the longest to then find the solution.

I have noticed that by giving as input an array generated with very similar values the execution time increases, but this is not always the case perhaps there are other factors involved.
Can I get some advice are there any conditions known to be more difficult? are they related to the branch and bound algorithm?