2. What is the most commonly used
class in java projects?
We use jmap tool from jdk to find the answer.
jmap -histo <java_app_pid> | head -n 14
Where <java_app_pid> - java application
process id (I have used pid of running tomcat
server)
3. The output of this command is:
num #instances #bytes class name
----------------------------------------------
1: 43101 18737736 [B
2: 61736 10054912 [C
3: 11598 5165024 [I
4: 58901 1413624 java.lang.String
5: 8514 749232 java.lang.reflect.Method
6: 21313 682016 java.util.HashMap$Node
7: 5286 549960 java.lang.Class
8: 7914 483944 [Ljava.lang.Object;
9: 9323 372920 java.util.HashMap$ValueIterator
10: 1814 320184 [Ljava.util.HashMap$Node;
11: 7862 314480 java.lang.ref.Finalizer
As we can see String is one of the common used classes in java projects and
takes a lot of memory. The most common operations performed with strings
is concatenation.
4. Strings
There are three variants of string concatenation:
1. String a = “Hello ” + “world”;
2. String b = new StringBuffer();
b.append(“Hello ”);
b.append(“world”).toString();
3. String c = “Hello ”.concat(“world”);
Let's look to them closer and try to compare...
5. stringA + stringB
There are two common possibilities to concatenate strings using '+'
sign:
1. String s1 = “STRING_VAL1” + “STRING_VAL2”;
2. String s2 = “STRING_VAL1” + STRING_VARIABLE;
We need to view generated bytecode by javap from jdk to
compare them:
javap -c SomeJava.class
6. For the first expression it looks like this.
Source:
String q = "Hello " + "world";
Bytecode:
public static void main(java.lang.String[]) throws java.lang.Exception;
Code:
0: ldc #3 // String Hello world
2: astore_1
3: return
7. For the second expression.
Source:
String hello = "Hello ";
String result = hello + "world";
Bytecode:
public static void main(java.lang.String[]) throws java.lang.Exception;
Code:
0: ldc #3 // String Hello
2: astore_1
3: new #4 // class java/lang/StringBuilder
6: dup
7: invokespecial #5 // Method java/lang/StringBuilder."<init>":()V
10: aload_1
11: invokevirtual #6 // Method java/lang/StringBuilder.append
14: ldc #7 // String world
16: invokevirtual #6 // Method java/lang/StringBuilder.append
19: invokevirtual #8 // Method java/lang/StringBuilder.toString
22: astore_2
23: return
8. If you try to concatenate two or more constant
strings then java compiler do this operation in
compile time:
String s = “Hello ” + “world” + “...”;
In other cases will be used explicitly or implicitly
StringBuilder and String.concat() method.
StringBuffer is also possible for multithreading
applications, but we examine only first two in bold.
Let's decide which of them better
9.
10. We will create two microbenchmark tests to find
out which is better StringBuilder or String.concat:
@State(Scope.Thread)
@OutputTimeUnit(TimeUnit.NANOSECONDS)
public class StringOpts {
@Benchmark
@BenchmarkMode({Mode.AverageTime})
@OutputTimeUnit(TimeUnit.MICROSECONDS)
public void concat_() {
String s1 = "Hello";
String s2 = s1.concat(" world");
}
@Benchmark
@BenchmarkMode({Mode.AverageTime})
@OutputTimeUnit(TimeUnit.MICROSECONDS)
public void append_() {
StringBuilder s1 = new StringBuilder("Hello");
String s2 = s1.append(" world").toString();
}
}
11. When benchmark had been finished we got such
result:
# Run complete. Total time: 00:00:12
Benchmark Mode Samples Score Error Units
t.StringOpts.append_ avgt 1 0.007 ± NaN us/op
t.StringOpts.concat_ avgt 1 0.010 ± NaN us/op
Where you can see that StringBuilder.append is
faster than String.concat function. When you need
concatenate only two strings in some cases you
can prefer String.concat because this generates
less objects than when you use StringBuilder.
12. String comparison
Comparison is the second common operation with
strings in java. Let's find out which of this four
functions are faster than other:
1. String.intern()
2. String.equals()
3. String.equalsIgnoreCase()
4. String.compareTo()
13. I decide to use in my benchmark array of 1000
random strings and compare elements from first
path (0..499) with elements from second path
(500..999):
@Setup
public void prepare() {
testStringsPool = new String[1000];
for (int i = 0; i < testStringsPool.length; i++) {
int customLength = rnd.nextInt();
if (customLength < 0) {
customLength *= -1;
}
testStringsPool[i] = randomString(customLength % 20 + 10);
}
}
14. Here are my benchmark functions without
annotations:
public void intern_() {
for (int i = 0; i < testStringsPool.length / 2; i++) {
if (testStringsPool[i].intern() ==
testStringsPool[testStringsPool.length - i – 1].intern());
}
}
public void equals_() {
for (int i = 0; i < testStringsPool.length / 2; i++) {
if (testStringsPool[i].equals(
testStringsPool[testStringsPool.length - i – 1]));
}
}
15. public void compareTo_() {
for (int i = 0; i < testStringsPool.length / 2; i++) {
if (testStringsPool[i].compareTo(
testStringsPool[testStringsPool.length - i - 1]) == 0);
}
}
public void equalsIgnoreCase_() {
for (int i = 0; i < testStringsPool.length / 2; i++) {
If(testStringsPool[i].equalsIgnoreCase(
testStringsPool[testStringsPool.length - i - 1]));
}
}
16. When we run those benchmark tests, we get
something similar like this:
# Run complete. Total time: 00:00:25
Benchmark Mode Samples Score Error Units
t.StringOpts.compareTo_ avgt 1 0.270 ± NaN us/op
t.StringOpts.equalsIgnoreCase_ avgt 1 1.323 ± NaN us/op
t.StringOpts.equals_ avgt 1 0.348 ± NaN us/op
t.StringOpts.intern_ avgt 1 148.612 ± NaN us/op
The winner is String.compareTo function
17. Conclusion for string comparison
1. compareTo – is the fastest because it
operates with parameter of String class without
additional checking for type safety
2. equals – a bit slower by checking input
parameter for the same type (String)
3. equalsIgnoreCase – more slower because all
the characters are converted to uppercase in
both strings
4. intern – the slowest. But when you need to
work with many identical strings it can help you
to reduce memory usage.