示例#1
0
  /*
  	- This reads the file and hashses the document, which are then stored in our arrayLisrs
  	- we do this before, so we dont have to hash again later ( which is time consuming)
  */
  private static void preliminaryStep(String dir) throws Exception {
    int start = 0; // start of the sliding window
    int end = start + window - 1; // ending boundary
    // prepoccessing step to hash the document, since we dont need to hash the document again
    for (int i = 0; i < fileList.size(); ++i) {
      // System.out.println("preliminaryStep " + fileList.get(i));
      Path p = Paths.get(dir + fileList.get(i)); // read this file
      byte[] array = Files.readAllBytes(p); // read the file in bytes
      // System.out.println(array.length);

      ArrayList<Long> md5Hashes = new ArrayList<Long>(); // make a new arrayList for this document
      HashDocument.hashDocument(
          array, md5Hashes, start,
          end); // this hashes the entire document using the window and stores itto md5hashes array

      // add the fileArray and hashedFile to our lists so we can use them later to run the
      // algorithms
      // note we hash and read file before, so we don't have to do it again
      fileArray.add(array);
      hashed_File_List.add(md5Hashes);
    }
  }
示例#2
0
  /*
  	-- This is a helper method to run the periodic dataset basically

  */
  private static void runPeriodic() throws Exception {
    System.out.println("Running TDDD Periodic");
    // this is alll the directories we will be running
    int arr[] = {10, 15, 20, 25, 30}; // this is the input number we will be running on
    // this is the base of the two files
    // these two are directories, we will concanate with the numbers to get the full dir name
    String base_old_file = "../../thesis-datasets/input_";
    String base_new_file = "../../thesis-datasets/periodic_";

    int total_iter_count =
        0; // this is used check how many times we will iterate through the data so we can make an
           // array of that size
    for (int i = startBoundary; i <= endBoundary; i += increment) total_iter_count++;

    for (int dir_num : arr) {
      // set up our directories

      String old_file_dir = base_old_file + dir_num + "/";
      String new_file_dir = base_new_file + dir_num + "/";
      System.out.println(old_file_dir);

      // read all the files in these two directories in sorted order
      ArrayList<String> old_file_list = new ArrayList<String>();
      ArrayList<String> new_file_list = new ArrayList<String>();

      ReadFile.readFile(old_file_dir, old_file_list);
      ReadFile.readFile(new_file_dir, new_file_list);

      // used to store all the runnings for the periodic data
      double[] block_size_list = new double[total_iter_count];
      double[] ratio_size_list = new double[total_iter_count];
      int totalRuns = 0;

      for (int i = 0; i < old_file_list.size(); ++i) {
        // System.out.println(old_file_list.get(i) + " " + new_file_list.get(i));
        String[] s1 = old_file_list.get(i).split("_");
        String[] s2 = new_file_list.get(i).split("_");
        // input file should corrospond to the output file
        if (!s1[1].equals(s2[1]) || !s1[2].equals(s2[2]))
          System.out.println("We got a huge problem");

        // basically same code as in the prelinaryStep method, but we need to modify it for
        // perdiodic files
        int start = 0; // start of the sliding window
        int end = start + window - 1; // ending boundary
        // we cant call preliminary function, so hash the two files individually
        // System.out.println("preliminaryStep " + fileList.get(i));
        Path p =
            Paths.get(
                old_file_dir
                    + old_file_list.get(
                        i)); // read the old file ( the one which we will be using as the base
                             // comparason)
        Path p2 =
            Paths.get(
                new_file_dir
                    + new_file_list.get(
                        i)); // read the old file ( the one which we will be using as the base
                             // comparason)

        byte[] old_file = Files.readAllBytes(p); // read the file in bytes
        byte[] new_file = Files.readAllBytes(p2);
        // System.out.println(array.length);
        ArrayList<Long> old_file_hashes =
            new ArrayList<Long>(); // make a new arrayList for this document
        ArrayList<Long> new_file_hashes =
            new ArrayList<Long>(); // make a new arrayList for this document

        HashDocument.hashDocument(
            old_file,
            old_file_hashes,
            start,
            end); // this hashes the entire document using the window and stores itto md5hashes
                  // array
        HashDocument.hashDocument(
            new_file,
            new_file_hashes,
            start,
            end); // this hashes the entire document using the window and stores itto md5hashes
                  // array

        // now call the startCdc method
        totalSize = new_file.length; // this is the length of the file
        startCDC(
            block_size_list, ratio_size_list, new_file, old_file, new_file_hashes, old_file_hashes);

        if (totalRuns % 10 == 0) System.out.println(totalRuns);
        totalRuns++;
      }

      // now output the results
      System.out.println("File dir = " + dir_num + " totalRuns = " + totalRuns);
      int index = 0;
      for (int i = startBoundary; i <= endBoundary; i += increment) {
        // avg out the outputs
        double blockSize = block_size_list[index] / (double) totalRuns;
        double ratio = ratio_size_list[index] / (double) totalRuns;
        System.out.println(i + " " + blockSize + " " + ratio);
        index++;
      }

      // now each index matches the corrosponding file
    }
  } // end of methid