Page 101 - ITU Journal Future and evolving technologies Volume 2 (2021), Issue 4 – AI and machine learning solutions in 5G and future networks
P. 101

ITU Journal on Future and Evolving Technologies, Volume 2 (2021), Issue 4




          First, find the main links whose end nodes are G or   on  the  network  service  rules,  rules  for  assessing
          H via the DFS algorithm [20,21], and save all such   node levels [22,23], and the requirements for link
          main links in the link library. For cross-connected   load balancing.
          main  links,  select  the  one  that  is  comparatively   max object_ratioavg – (Eavg+Emin+Emax) – α · sub_ratio
          longer  than  the  others  in  accordance  with      -β · hang_ratio                             (1)
          constraint rules on cross-connected links.
                                                               Explain:
          Regarding  the  sub-links  and  hanging  links,  the
          innovative "Node-Removing Method" is applied to      object_ratioavg:  Daily  average  of  the  ratio  of  links
          sort  them  out  when  all  main  links  have  been   whose  bandwidth  utilization  is  optimized  to  the
          identified. Specifically speaking, after main links in   target range per hour
          the network topology are removed, the related sub-   Eavg:  Daily  average  of  the  E-value  per  hour  that
          links  and  hanging  links  will  be  disconnected   indicates network load balancing
          accordingly. In this way, all the links in this network
          topology can be sorted out by category, and all the   Emin: Minimum E-value per hour of each day
          links of the topology will be recorded in the library   Emax: Maximum E-value per hour of each day
          by category.
                                                               sub_ratio: Proportion of nodes on sub-links to those
          1)  Select one main link from the link library, and   on the entire link (daily average of values per hour)
              then remove it by using NetworkX.                hang_ratio: Proportion of nodes on hanging links to
          2)  Execute  the  connected_components()  method     those on the entire links (daily average of values per
              to identify all connected sub-graphs [15].       hour)
          3)  Review  all  these  sub-graphs  and  remove  the   α,β: Constant coefficients that are set to 0.02  and
              sub-graphs that are not related to the current   0.05 respectively.
              main link in accordance with the existing node   Edge addition: Distance between two nodes that are
              connections.
                                                               to  be  connected  by  adding  an  edge  for  topology
          4)  Associate the sub-links and hanging links in the   optimization should be within 500 meters.
              sub-graphs that have been removed in step 3      Number of nodes on each link: Maximally 30 nodes
              with  the  correct  main  link,  and  update  the   on each link.
              collection of neighbors’ links in the link library.
          5)  Continue  to  remove  other  main  links  in  the   Target  range  of  link  bandwidth  utilization:
                                                                3
                                                                      7
              same  way  until  all  the  sub-links  and  hanging   [     ,        ]
                                                                5
                                                                      5
              links  in  the  graph  are  associated  with
              appropriate  main  links,  as  shown  in  Fig.  6   Formula used to calculate E-value:
              below.                                                          ∑                                 
                                                                                 
                                                                            =                              (2)
                                                                            
                                                                                   A
                                                               "E" is the variance of the link bandwidth utilization
                                                               (fp) of all links in the network topology.
                                                               "μ" is the average of bandwidth utilization of all the
                                                               links in the network topology. In the formula, fx is
                                                               the traffic on the nodes (excluding the start and end
                                                               nodes) on the link, and A stands for the maximum
                                                               node  capacity  that  one  node  (excluding  the  start
                                                               and end nodes) on the link can possess.
                                                               3.4  Build  and  complete  network  topology
                                                                     optimization  and  find  the  optimal

                     Fig. 6 – Node removing method                   solution through iteration
                                                               Network topology optimization for load balancing can
          3.3  Evaluation of network topology
                                                               be implemented in three different ways [24,25,26,27],
          The following formula describes how the effect of    including  link  combination,  partial  link  optimization,
          network topology optimization is evaluated based     and the optimization by node transfer.






                                             © International Telecommunication Union, 2021                    85
   96   97   98   99   100   101   102   103   104   105   106