Cross Section
One of the JavaIdioms.
The memory allocated for each Java object includes a header used by the JVM.
This header includes the per-object mutex, a reference to the object's class and other information used by the heap and runtime.
When allocating a large array of ValueObjects, the memory used by the headers can be a significant overhead if each object in the array is small.
An example would be:
class Point2DList {
static private final int POINT_COUNT = 10000;
private Point2D[] _points;
public Point2DList() {
_points = new Point2D[POINT_COUNT];
for( int i = 0; i < POINT_COUNT; i++ ) {
_points[i] = new Point2D.Double(0.0,0.0);
}
}
public Point2D getPoint( int n ) {
return _points[n];
}
public void setPoint( int n, Point2D p ) {
_points[n] = p;
}
}
The object header used by the HotSpot JVM is 8 bytes long and a double is stored in 8 bytes. Therefore the amount of memory allocated for each Point2D object is 24 bytes and the overhead for each object is 1/3 of its size!
Therefore: allocate multiple "cross section" arrays to hold what would be the instance variables of each logical object in the collection. Hide these arrays behind a FacadePattern that makes them appear to be a single array of objects that each have multiple instance variables.
For instance, the Point2DList example becomes:
class Point2DList {
static private final int POINT_COUNT = 10000;
private double[] _x; // The X "cross section"
private double[] _y; // The Y "cross section"
public Point2DList(){
_x = new double[POINT_COUNT];
_y = new double[POINT_COUNT];
for( int i = 0; i < POINT_COUNT; i++ ) {
_x[i] = 0.0;
_y[i] = 0.0;
}
}
public Point2D getPoint( int n ) {
return new Point2D.Double( _x[n], _y[n] );
}
public void setPoint( int n, Point2D p ) {
_x[n] = p.getX();
_y[n] = p.getY();
}
}
(This is a slightly contrived example because the x and y "cross sections" both have the double type and so could be stored in the same array, but you get the idea, I hope.)
Comments, known uses? --NatPryce
Known uses:
- I have used this pattern in a framework for animating line drawings. --NatPryce
- CrossSections are used to reduce the memory requirements of the LTSA model checker (http://www-dse.doc.ic.ac.uk/concurrency/).
- A CrossSection uses the FacadePattern
- CrossSection is a specific use of EvisceratedObjects
- The java.awt.Polygon class uses CrossSections.
-
Fortran made (makes?) heavy usage of this idiom because it doesn't have records.
- You'll find many algorithms are written up using CrossSection as well (due to Fortran's influence, no doubt).
- dimensional databases use this scheme--RichardHenderson.
Alternatives:
- Languages like C++, Eiffel and C# have value/expanded objects which can be put in arrays without the overhead, so they don't need this pattern.
As with all optimizations, never apply unless you've proven (with measurements, not guesses) that you need it.
Is this pattern common in Smalltalk?