Object pool managers
Object pool managers control the reuse of application objects and Developer Kit objects, such as Vectors and HashMaps.
Multiple object pool managers can be created in an Application Server cell. Each object pool manager has a unique cell-wide JNDI name. Applications can find a specific object pool manager by doing a JNDI lookup using the specific JNDI name.
The object pool manager and its associated objects implement the following interfaces:
public interface ObjectPoolManager { ObjectPool getPool(Class aClass) throws InstantiationException, IllegalAccessException; ObjectPool createFastPool(Class aClass) throws InstantiationException, IllegalAccessException; } public interface ObjectPool { Object getObject(); void returnObject(Object o); }
The getObject() method removes the object from the object pool. If a getObject() call is made and the pool is empty, then an object of the same type is created. A returnObject( ) call puts the object back into the object pool. If returnObject() is not called, then the object is no longer allocatable from the object pool. If the object is not returned to the object pool, then it can be garbage collected. Each object pool manager can be used to pool any Java object with the following characteristics:
Each pooled object class must have its own object pool. In addition, an application gets an object pool for a specific object using either the ObjectPoolManager.getPool() method or the ObjectPoolManager.createFastPool() method. The difference between these methods is that the getPool() method returns a pool that can be shared across multiple threads. The createFastPool() method returns a pool that can only be used by a single thread.
If in a JVM, the getPool() method is called multiple times for a single class, the same pool is returned. A new pool is returned for each call when the createFastPool() method is called. Basically, the getPool() method returns a pool that is thread-synchronized. The pool for use by multiple threads is slightly slower than a fast pool because of the need to handle thread synchronization. However, extreme care must be taken when using a fast pool. Consider the following interface:
public interface PoolableObject
{
void init();
void returned();
}
If the objects placed in the pool implement this interface and the ObjectPool.getObject() method is called, the object that the pool distributes has the init() method called on it. When the ObjectPool.returnObject() method is called, the PoolableObject.returned() method is called on the object before it is returned to the object pool. Using this method objects can be pre-initialized or cleaned up. It is not always possible for an object to implement PoolableObject. For example, an application might want to pool ArrayList objects. The ArrayList object needs clearing each time the application reuses it. The application might extend the ArrayList object and have the ArrayList object implement a poolable object. For example, consider the following:
public class PooledArrayList extends ArrayList implements PoolableObject
{
public PooledArrayList()
{
}
public void init() {
}
public void returned()
{
clear();
}
}
If the application uses this object, in place of a true ArrayList object, the ArrayList object is cleared automatically when it is returned to the pool.
Clearing an ArrayList object simply marks it as empty and the array backing the ArrayList object is not freed. Therefore, as the application reuses the ArrayList, the backing array expands until it is big enough for all of the application requirements. When this point is reached, the application stops allocating and copying new backing arrays and achieves the best performance.
It might not be possible or desirable to use the previous procedure. An
alternative is to implement a custom object pool and register this pool with
the object pool manager as the pool to use for classes of that type. The class
is registered by the WebSphere administrator when the object pool manager
is defined in the cell. Take care that these classes are packaged in Java
Archive (JAR) files available on all of the nodes in the cell where they might
be used.