Will programs get faster just by increasing the Java version?
(This article is a multipost with Go to the Horizon)
I often hear that increasing the running version of Java makes it faster. But is that really the case? And how fast would it be if it were true?
So, I experimented with a simple program.
Experiment outline
For the experiment, I created Java program to solve Sudoku. Since this program only repeats simple operations, the tendency is probably different from that of a complicated program such as a Web application, but I think it will be a reference level.
Compile this with ** Java 1.1 ** and ** with each Oracle JDK (32bit / 64bit) from Java 1.1 to 12 ** Dataset of 1 million questions in Sudoku .com / bryanpark / sudoku / version / 3) was read and the time to complete the solution was measured. [^ compile] **
[^ compile]: Because it is backwards compatible, class files compiled with Java 1.1 will continue to work in later versions.
The detailed measurement conditions are as follows.
- Execution environment
- Windwos 10 Home 1809 (64bit)
- Intel Core i7-7500 CPU @ 2.70GHz
- Java execution options
- Grant
-mx512m`` -ms512m[^ jvmoption] - Since Java 1.3.1, add
-server[^ server] - Measuring method
- Measure 5 times for each version and get the median
[^ jvmoption]: You might wonder, "That's -Xmx -Xms?", But in Java 1.1 this option is -mx`` -ms. In Java 1.2, the VM option was changed to -Xmx`` -Xms, but you can still use -mx`` -ms.
[^ server]: Without this option, the HotSpot Client VM would be used, which would be about twice as slow ...
Experimental result
Since the processing is divided into two parts, the graph axis is separated for each processing.
- Processing to read a data set of 1 million questions in Sudoku * (blue part in the figure below) *
- This part * depends mainly on the performance of the GC **
- Processing to solve Sudoku * (red part in the figure below) *
- This part depends mainly on the performance of JIT optimization **
Here are the results of the experiment.
Original data of graph ...
Java
Version
Data read(ms)
Sudoku analysis(ms)
total(ms)
Java 1.1 (32bit)
1.1.8
8,313
4,231
12,544
Java 1.2 (32bit)
1.2.2_017
6,647
4,360
11,006
Java 1.3 (32bit)
1.3.0_05
10,747
6,905
17,652
Java 1.3.1 (32bit)
1.3.1_28
7,647
5,093
12,740
Java 1.4 (32bit)
1.4.0_04
8,061
4,741
12,802
Java 1.4.1 (32bit)
1.4.1_07
7,810
4,796
12,606
Java 1.4.2 (32bit)
1.4.2_19
3,437
3,281
6,717
Java 5 (32bit)
1.5.0_22
3,050
3,289
6,338
Java 6 (32bit)
1.6.0_45
3,530
4,077
7,607
Java 7 (32bit)
1.7.0_80
3,411
3,211
6,622
Java 8 (32bit)
1.8.0_202
3,239
3,257
6,496
Java 6 (64bit)
1.6.0_45
5,029
3,312
8,341
Java 7 (64bit)
1.7.0_80
7,019
3,148
10,167
Java 8 (64bit)
1.8.0_202
5,050
3,054
8,104
Java 9 (64bit)
9.0.4
2,101
2,961
5,061
Java 10 (64bit)
10.0.2
2,056
3,054
5,110
Java 11 (64bit)
11.0.3
1,922
3,343
5,265
Java 12 (64bit)
12.0.1
1,934
3,422
5,356
| Java | Version | Data read(ms) | Sudoku analysis(ms) | total(ms) |
|---|---|---|---|---|
| Java 1.1 (32bit) | 1.1.8 | 8,313 | 4,231 | 12,544 |
| Java 1.2 (32bit) | 1.2.2_017 | 6,647 | 4,360 | 11,006 |
| Java 1.3 (32bit) | 1.3.0_05 | 10,747 | 6,905 | 17,652 |
| Java 1.3.1 (32bit) | 1.3.1_28 | 7,647 | 5,093 | 12,740 |
| Java 1.4 (32bit) | 1.4.0_04 | 8,061 | 4,741 | 12,802 |
| Java 1.4.1 (32bit) | 1.4.1_07 | 7,810 | 4,796 | 12,606 |
| Java 1.4.2 (32bit) | 1.4.2_19 | 3,437 | 3,281 | 6,717 |
| Java 5 (32bit) | 1.5.0_22 | 3,050 | 3,289 | 6,338 |
| Java 6 (32bit) | 1.6.0_45 | 3,530 | 4,077 | 7,607 |
| Java 7 (32bit) | 1.7.0_80 | 3,411 | 3,211 | 6,622 |
| Java 8 (32bit) | 1.8.0_202 | 3,239 | 3,257 | 6,496 |
| Java 6 (64bit) | 1.6.0_45 | 5,029 | 3,312 | 8,341 |
| Java 7 (64bit) | 1.7.0_80 | 7,019 | 3,148 | 10,167 |
| Java 8 (64bit) | 1.8.0_202 | 5,050 | 3,054 | 8,104 |
| Java 9 (64bit) | 9.0.4 | 2,101 | 2,961 | 5,061 |
| Java 10 (64bit) | 10.0.2 | 2,056 | 3,054 | 5,110 |
| Java 11 (64bit) | 11.0.3 | 1,922 | 3,343 | 5,265 |
| Java 12 (64bit) | 12.0.1 | 1,934 | 3,422 | 5,356 |
Consideration
The premise is "in this program", but the following can be seen from the data.
- Compared to Java 1.1, modern Java is more than twice as fast
- It's slowing down sharply in Java 1.3. Was HotSpot not mature yet?
- It was slow until Java 1.4.1, but much faster with Java 1.4.2
- Is the data loading faster because the default GC has been switched to CMS [^ cms]? [^ skrb]
- Use of SSE and SSE2 instructions on IA32 platforms in the release notes (https://web.archive.org/web/20110320141919/http://www.oracle.com/technetwork/java/javase/changes-136374. Since it says html # vm), is this the reason why the math analysis is faster?
- Java 6-8 reads data faster in 32bit than in 64bit VM
- Much faster with Java 9
[^ cms]: Concurrent Mark Sweep Garbage Collector. For more information on this GC, see KUBOTA Yuji's Material.
Summary
** Just increase the Java version and your program will be faster. ** **
Recommended Posts