org.javaseis.parallel
Class ParallelMap

java.lang.Object
  org.javaseis.parallel.ParallelMap

All Implemented Interfaces:

java.io.Serializable

public class ParallelMap
extends java.lang.Object
implements java.io.Serializable

Wrap a Java Collections Map with parallel synchronization semantics.

Use java.util.Map as a repository for arbitrary collections of serializable objects. MPIJava can then be used to communicate the map contents via broadcast, send, and merge operations.

The magic of MPI and Java Serialization make it all work. A "master" task can create a map and fill it with a variety of serializable objects:

 IParallelContext pc = new MPIContext();
 ParallelMap pmap = new ParallelMap;
 if (pc.isMaster()) {
    DataContext dataContext = new DataContext( args );
    Coordinates xyz = new Coordinates( 0, 0, 0 );
    Integer rank = new Integer( pc.rank() );
    Integer size = new Integer( pc.size() );
    pmap.put("DataContext", dataContext );
    pmap.put("Coordinates", xyz );
    xyz.setValues( 1, 1, 1 );
    pmap.put("Size", size );
    pmap.put("Rank", rank );
 }
 pmap.broadcast();
 Integer rank = pmap.get("Rank");
 pc.serialPrint("Rank after broadcast = " + rank.intValue() );
 pc.serialPrint("XYZ Coorinates = " + xyz.toString() );
 

In the above code fragment, all tasks create a parallel map. The master task fills it with a variety of serializable objects. In the broadcast, the master task serializes the map, which in turn serializes every object in the map. The serialized data stream is sent to all tasks in the parallel context. On the receiving end, the map is "deserialzed". As each object in the map is received it is created and placed in the map. Note that the receiving task does not need to allocate anything other than the map. Of course the receiving task has to be able to provide a matching object type when an object is retrieved from the map. In the example above, the retrieved rank will be 0, as received from the master.

Note that the value of the Coordinates object will be (1,1,1) rather than (0,0,0) as originally placed in the map, since the "xyz.setValues" updated the object referred to by the map.

Typical usage is to maintain a synchronous parallel state within a for a set of tools in a processing stream. Two modes are currently supported. "broadcast" sends the contents of the map from a master node to all other tasks. "merge" collects the map from all nodes in rank order and merges the results into the map on the master task, followed by a broadcast that replaces the map on all tasks with the merged map.

Author:

Chuck Mosher for JavaSeis.org

See Also:

Serialized Form

Field Summary

static int

MAP_TAG

Constructor Summary

ParallelMap()
Create a new parallel map.

Method Summary

void	broadcast(IParallelContext pc) Broadcast the contents of the map to all tasks in the parallel context
java.lang.Object	get(java.lang.String key) Retrieve an object from the map
java.util.Map<java.lang.String,java.lang.Object>	getMap() Return the Map object for the ParallelMap
java.util.Set<java.lang.String>	keySet() Return the set of keys contained in the parallel map
static void	main(java.lang.String[] args)
void	merge(IParallelContext pc) Merge map entries from all tasks in a parallel context.
void	put(java.lang.String key, java.lang.Object value) Put an object into the parallel map
void	putAll(java.util.Map m)

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Detail

MAP_TAG

public static final int MAP_TAG

See Also:

Constant Field Values

Constructor Detail

ParallelMap

public ParallelMap()

Create a new parallel map. The current map implementation is "LinkedHashMap", which keeps the map entries in the order in which they are placed in the map.

Method Detail

broadcast

public void broadcast(IParallelContext pc)

Broadcast the contents of the map to all tasks in the parallel context

Parameters:

pc - parallel context to be used for the broadcast

get

public java.lang.Object get(java.lang.String key)

Retrieve an object from the map

Parameters:

key - the key value for the object

Returns:

the object that was stored

getMap

public java.util.Map<java.lang.String,java.lang.Object> getMap()

Return the Map object for the ParallelMap

Returns:

Map object

keySet

public java.util.Set<java.lang.String> keySet()

Return the set of keys contained in the parallel map

Returns:

key set for the map

main

public static void main(java.lang.String[] args)

Parameters:

args -

merge

public void merge(IParallelContext pc)

Merge map entries from all tasks in a parallel context. The maps are retrieved in rank order from each task, and the map on the master task is updated. The master task then broadcasts the map to all tasks, which first clear their local copy, and then replace the map with the merged version.

Parameters:

pc - parallel context for the merge

put

public void put(java.lang.String key,
                java.lang.Object value)

Put an object into the parallel map

Parameters:

key - the key to be used for retrieving the object

value - the serializable object to be placed in the map

putAll

public void putAll(java.util.Map m)