How to get comparable Oracle JDBC performance using Java 1.4 vs 1.1.7?

Question

Our application makes extensive use of JDBC to access an Oracle database. We wrote it a number of years ago using java 1.1.7 and we have been unable to move to new versions of java because of the performance degradation.
I traced the problem to JDBC calls. I can reproduce the problem using a simple program that simply connects to the database, executes a simple query and then reads the data. The same program running under java 1.4 is about 60% slower than under java 1.1.7. The query is about 80% slower and getting the data is about 150% slower!
The program is attached below. Note, I run the steps twice as the first time the times are much larger which I attribute to java doing some initializations. So the second set of values I think are more representative of the actual performance in our application where there are numerous accesses to the database. Specifically, I focus on step 4 which is the execute query command and step 5 which is the data retrieval step. The table being read has 4 columns with 170 tuples in it.
Here are the timings I get when I run this on a Sparc Ultra 5 running
SunOs 5.8 using an Oracle database running 8.1.7:
                 java 1.1.7  java 1.4
        overall:    2.1s         3.5s
        step 1:     30           200
        step 2:    886          2009
        step 3:      2             2
        step 4:      9            17
        step 5:    122           187
        step 6:      1             1
        step 1:      0             0
        step 2:    203           161
        step 3:      0             1
        step 4:      8            15   <-   87% slower
        step 5:     48           117   <-  143% slower
        step 6:      1             2I find the same poor performance from java versions 1.2 and 1.3.
I tried using DataDirect's type 4 JDBC driver which gives a little better performance but overall it is still much slower than using java 1.1.7.
Why do the newer versions of java have such poor performance when using JDBC?
What can be done so that we can have performance similar to java 1.1.7
using java 1.4?
========================================================================
import java.util.*;
import java.io.*;
import java.sql.*;
public class test12 {
    public static void main(String args[]) {
        try {
                long time1 = System.currentTimeMillis();
/* step 1 */  DriverManager.registerDriver(
                    new oracle.jdbc.driver.OracleDriver());
                long time2 = System.currentTimeMillis();
/* step 2 */  Connection conn = DriverManager.getConnection (
              "jdbc:oracle:thin:@dbserver1:1521:db1","user1","passwd1");
                long time3 = System.currentTimeMillis();
/* step 3 */  Statement stmt = conn.createStatement();
                long time4 = System.currentTimeMillis();
/* step 4 */  ResultSet rs = stmt.executeQuery("select * from table1");
                long time5 = System.currentTimeMillis();
/* step 5 */  while( rs.next() ) {
                  int message_num = rs.getInt(1);
                  String message = rs.getString(2);
                long time6 = System.currentTimeMillis();
/* step 6 */  rs.close(); stmt.close();
                long time7 = System.currentTimeMillis();
            System.out.println("step 1: " + (time2 - time1) );
            System.out.println("step 2: " + (time3 - time2) );
            System.out.println("step 3: " + (time4 - time3) );
            System.out.println("step 4: " + (time5 - time4) );
            System.out.println("step 5: " + (time6 - time5) );
            System.out.println("step 6: " + (time7 - time6) );
            System.out.flush();
        } catch ( Exception e ) {
            System.out.println( "got exception: " + e.getMessage() );
        ... repeat the same 6 steps again...
}

Answer 1

If I run my sample program with the -server option, it
takes a lot longer (6.8s vs 3.5s).Which has to be expected, as the -server option optimizes for long running programs - so it shoudl go with my second suggestion, more below...
I am not certain what you mean by "just let the jvm
running". Our users issue a command (in Unix) which
invokes one of our java programs to access or update
data in a database. I think what you are suggesting
would require that I rewrite our application to have a
java program always running on the users workstation
and to also rewrite our
commands (over a hundred) to some how pass data and
receive data with this new server java program. That
does not seem a very reasonable just to move to a new
version of java. Or are you suggesting something
else?No I was just suggestion what you descript. But if this is not an option, then maybe you should merge your java-programs to C or another native language. Or you could try the IBM-JDK with the -faststart (or similar) option. If thew Unix you mention is AIX, then there would be the option of a resetable-vm. But I cannot say if this VM would solve your problem. Java is definitly not good for applications which only issue some unqiue commands because the hotspot-compiler can not be efficiently used there. You can only try to get 1.1.7 performance by experimenting with vm-parameters (execute java -X).