The document discusses JVM performance options, including various flags and their configurations for garbage collection, performance tuning, and debugging. It details the impact of specific options like escape analysis, autoboxing, and optimized string concatenation on performance improvements. Additionally, it highlights memory optimization techniques such as compressed object pointers and large pages for enhancing JVM efficiency.

1. JVM performance
options.
How it works?

2. Dmitriy Dumanskiy
Cogniance, Velti project
Java Team Lead

3. Xmx2048M -Xms2048M
-XX:ParallelGCThreads=8 -Xincgc
-XX:+UseConcMarkSweepGC -XX:
+UseParNewGC -XX:+CMSIncrementalPacing
-XX:+AggressiveOpts
-XX:+CMSParallelRemarkEnabled
-XX:+DisableExplicitGC
-XX:MaxGCPauseMillis=500
-XX:SurvivorRatio=16
-XX:TargetSurvivorRatio=90
-XX:+UseAdaptiveGCBoundary
-XX:-UseGCOverheadLimit -Xnoclassgc
-XX:UseSSE=3 -XX:PermSize=128m
-XX:LargePageSizeInBytes=4m

5. Options may vary per
architecture / OS / JVM version

6. JVM 6 ~ 730 options
JVM 7 ~ 680 options

7. -X : are non-standard (not all JVM)
-XX : are not stable

8. Types
Boolean : -XX:+<option> or -XX:-<option>
Numeric : -XX:<option>=<number>
String : -XX:<option>=<string>

9. Categories

Behavioral options

Garbage Collection options

Performance tuning options

Debugging options

10. -XX:+DoEscapeAnalysis
Analys :

Can objects be created on stack?

Are objects accessed from 1 thread?

11. -XX:+DoEscapeAnalysis
Analys result :

GlobalEscape

ArgEscape

NoEscape

12. NoEscape
class Cursor {
String icon;
int x;
public void create() {
Cursor c = new Cursor(); //HEAP
c.icon = null; //HEAP
c.x = 0; //HEAP
}
}

13. NoEscape → scalar replacement
class Cursor {
String icon;
int x;
public void create() {
String icon = null; //ref on stack frame
int x = 0; //int on stack frame
}
}

14. NoEscape → scalar replacement

15. NoEscape → scalar replacement

16. -XX:+DoEscapeAnalysis
~20-60% locks elimination
~15-20% performance improvement

17. -XX:+DoEscapeAnalysis

18. -XX:+AggressiveOpts
-AggressiveOpts +AggressiveOpts
AutoBoxCacheMax 128 20000
BiasedLockingStartupDelay 4000 500
EliminateAutoBox false true
OptimizeFill false true
OptimizeStringConcat false true

19. -XX:AutoBoxCacheMax=size
Sets IntegerCache.high value :
class Integer {
public static Integer valueOf(int i) {
if(i >= -128 && i <= IntegerCache.high)
return IntegerCache.cache[i + 128];
else
return new Integer(i);
}
}

20. -XX:AutoBoxCacheMax=size
new Integer(1) vs Integer.valueOf(1)
valueOf ~4 times faster

21. -XX:BiasedLockingStartupDelay=delay

Biased

Thin

Fat

22. -XX:-OptimizeStringConcat
String twenty = «12345678901234567890»;
String sb = twenty + twenty + twenty +
twenty;
String twenty = «12345678901234567890»;
String sb = new StringBuilder()
.append(twenty).append(twenty)
.append(twenty).append(twenty).toString();

23. -XX:-OptimizeStringConcat
String twenty = «12345678901234567890»;
String sb = new StringBuilder()
.append(twenty).append(twenty)
.append(twenty).append(twenty).toString();
new char[16];
new char[34];
new char[70];
new char[142];

24. -XX:+OptimizeStringConcat
String twenty = «12345678901234567890»;
String sb = new StringBuilder()
.append(twenty).append(twenty)
.append(twenty).append(twenty).toString();
new char[80];

25. -XX:+OptimizeStringConcat
String twenty = «12345678901234567890»;
StringBuilder sb1 = new StringBuilder();
sb1.append(new StringBuilder()
.append(twenty).append(twenty)
.append(twenty).append(twenty)
);
new char[80];

26. XX:+OptimizeFill
Arrays.fill(), Arrays.copyOf() or code
patterns :
for (int i = fromIndex; i < toIndex; i++) {
a[i] = val;
}
Native machine instructions

27. XX:+EliminateAutoBox

Removes unnecessary AutoBox operations

Works only for Integers

28. -XX:+UseStringCache
Look like not used anymore

29. -XX:+UseCompressedStrings
For ASCII characters:
char[] -> byte[]

30. -XX:+UseCompressedOops

Heap size up to 32Gb

References size 50% smaller

JVM performance boost 2-10%

20 — 60% less memory consumption;

31. -XX:+UseCompressedOops
1.2
1
0.8
0.6
0.4
0.2
0
32-bit 64-bit 64-bit Comp.

32. -XX:+EliminateLocks
synchronized (object) {
synchronized (object) {
//doSomething1
//doSomething1
}
//doSomething2
synchronized (object) {
}
//doSomething2
}

33. -XX:+EliminateLocks
synchronized (object) {
//doSomething1
} synchronized (object) {
//doSomething1
//doSomething2 //doSomething2
//doSomething3
synchronized (object) { }
//doSomething3
}

34. -XX:+UseLargePages
Translation-Lookaside Buffer
(TLB) is a page translation
cache that holds the most-
recently used virtual-to-physical
address translations

35. -XX:CompileThreshold=n
Client mode n = 1500
Server mode n = 10000
More profile data — more optimizations

36. -XX:hashCode=n
Object.hashCode() - internal
address of the object?

37. -XX:hashCode=n
n is :

0 – Park-Miller RNG (default)

1 – f (address, global state)

2 – const 1

3 – sequence counter

4 – object address

5 – Thread-local Xorshift