MODELING ABSTRACT NETWORKING FILTER WITH THE ABILITY TO CLASSIFY PEER-TO-PEER INTERACTIONS

Abstract

Develop  a  mathematical  model  of  an  abstract  network  packet  filter  with  the  ability  to classify Peer-to-Peer interactions. Used  methods  of  mathematical  modeling,  simulation  modeling  for  the  method  of group method of data handling and methods of mathematical statistics. Verification of effectiveness of the proposed models and methods is performed by comparing various metrics of the classifier of peer-to-peer interaction.  The  combination  of  different  approaches  in  the  synthesis  of  network  filter  rules  allows  us  to abstract  from the  transport layer  protocols, the  rules  are  described  as  a binary  tree that is  searched for peer-to-peer interaction properties.  The  paper  proposes  a  mathematical  model  of  an  abstract  network  packet  filter,  which allows the use of a flexible accounting policy in networks of general purpose. Under the accounting term we meant  the  possibility  of  creating  restrictions  for  a  number  of  users  who  create  the  largest  volumes  of information  flows,  thereby  affecting  other  participants  in  the  network  segment.  The  establishment  of  an effective  procedure  to  combat  this  phenomenon  will  improve  the  quality  of  the  services  provided  and minimizes the possibility of exceeding the level of allowed bandwidth. The use of an abstract network filter can be combined with a system for monitoring proper work of a multiservice network, thereby providing a systematic approach in identifying problems and violations of access policies. Scientific  originality.  The  proposed  model  by  combination  of  the  reviewed  methods  allow  us  to
identify peer-to-peer interaction with increased accuracy. Particular importance is vital as the process of creating  classification  filter  rules,  permits  to  use  external  tools  that  provide  interaction  signatures  of applications. Practical value. The results of theoretical studies were implemented as a separate software module of  the  packet  classification  system  for  automatically  determining  the  parameters  of  interaction  between applications  in  a  peer-to-peer  network.  The  learning  process  of  the  classification  network  is  carried  out automatically, which results in complete autonomy of the system.