5 Replies Latest reply on Nov 10, 2006 4:43 PM by clebert.suconic

    Poor performance with test...

    davidecr

      trying this...

      long init = System.currentTimeMillis();
      ObjectOutputStream dos = new JBossObjectOutputStream(new FileOutputStream("c:\\test.txt"));
      dos.writeObject(opportunities);
      dos.close();
      File f = new File("c:\\test.txt");
      System.out.println("result bytes = " + f.length());
      System.out.println("result time = " + (System.currentTimeMillis() - init));

      opportunities is an ArrayList filled with a javabean with no superclasses an using as fields primitives string and java.util.Date

      with JBossObjectOutputStream has the result...
      INFO: result bytes = 125880 (Constant)
      INFO: result time = 828 (%)


      with java.io.ObjectOutputStream has the result...
      INFO: result bytes = 103723 (103723)
      INFO: result time = 156 (%)

      I'm doing something wrong??? as I don;t getting even the same performanceas normal serialization.

        • 1. Re: Poor performance with test...
          clebert.suconic

          You always need a BufferedOutputStream and BufferedINputStream on JBossObjectOutputStream...

          even the regular ObjectOutputStream will work better with a BufferedStream

          There is another thread on this forum talking about this.


          You should do:



          long init = System.currentTimeMillis();
          ObjectOutputStream dos = new JBossObjectOutputStream (new BufferedOutputStream(new FileOutputStream("c:\\test.txt")));
          dos.writeObject(opportunities);
          dos.close();
          File f = new File("c:\\test.txt");
          System.out.println("result bytes = " + f.length());
          System.out.println("result time = " + (System.currentTimeMillis() - init));

          • 2. Re: Poor performance with test...
            lxk123

            I have similar experience: this package is slower than standard JDK 1.4.2 and consuming more memory. Is there something wrong on the following code snippet?


            ---------------------
            ByteArrayOutputStream bstream = new ByteArrayOutputStream();
            org.jboss.serial.io.JBossObjectOutputStream out = new org.jboss.serial.io.JBossObjectOutputStream(bstream);

            out.writeObject(obj);
            out.close();

            // Loading the memory block as an Object
            ByteArrayInputStream binstream = new ByteArrayInputStream(bstream.toByteArray());
            org.jboss.serial.io.JBossObjectInputStream oin = new org.jboss.serial.io.JBossObjectInputStream(binstream);

            Object newobj = oin.readObject();

            • 3. Re: Poor performance with test...
              clebert.suconic

              You are probably measuring classLoading time...

              You need to ignore first measuring... considering only steady time.

              • 4. Re: Poor performance with test...
                lxk123

                The measure is taking against 1000 runs of cloning a same object, and the class loading should not be a big impact here.

                By the way, the smartClone method provided works great. On my test, it is more than twice as fast as the standard serialization.

                -------------
                ByteArrayOutputStream bstream = new ByteArrayOutputStream();
                org.jboss.serial.io.JBossObjectOutputStream out = new org.jboss.serial.io.JBossObjectOutputStream(bstream);

                Object newobj = out.smartClone(obj);

                • 5. Re: Poor performance with test...
                  clebert.suconic

                  On the Benchmarks I have done I always ignored 100 first elements to be fair.. not computing ClassLoading as JDK classes get sometimes pre-loaded (Specially Serialization)

                  Also.. I always used the same ByteArrrayOutputStream (by calling reset)...

                  And of course my Java and JBoss examples were the same. I needed to be fair on both sides.