com.bigdata.rdf.graph.impl

Class GASEngine

java.lang.Object

com.bigdata.rdf.graph.impl.GASEngine

All Implemented Interfaces:

IGASEngine

Direct Known Subclasses:

BigdataGASEngine, RAMGASEngine, SAILGASEngine

public abstract class GASEngine
extends Object
implements IGASEngine

IGASEngine for dynamic activation of vertices. This implementation maintains a frontier and lazily initializes the vertex state when the vertex is visited for the first time. This is appropriate for algorithms, such as BFS, that use a dynamic frontier. TODO Algorithms that need to visit all vertices in each round (CC, BC, PR) can be more optimially executed by a different implementation strategy. The vertex state should be arranged in a dense map (maybe an array) and presized. For example, this could be done on the first pass when we identify a vertex index for each distinct V in visitation order. TODO Vectored expansion with conditional materialization of attribute values could be achieved using CONSTRUCT. This would force URI materialization as well. If we drop down one level, then we can push in the frontier and avoid the materialization. Or we can just write an operator that accepts a frontier and returns the new frontier and which maintains an internal map containing both the visited vertices, the vertex state, and the edge state. TODO Some computations could be maintained and accelerated. A great example is Shortest Path (as per RDF3X). Reachability queries for a hierarchy can also be maintained and accelerated (again, RDF3X using a ferrari index). TODO Option to materialize Literals (or to declare the set of literals of interest) [Note: We can also require that people inline all URIs and Literals if they need to have them materialized, but a materialization filter for Gather and Scatter would be nice if it can be selective for just those attributes or vertex identifiers that matter).

Author:

Bryan Thompson

Constructor Summary

Constructors

Constructor and Description
GASEngine(int nthreads)

Method Summary

Methods

Modifier and Type	Method and Description
ExecutorService	getGASThreadPool() If there is an ExecutorService for the GASEngine, then return it (nthreads GT 1).
int	getNThreads() The parallelism for the SCATTER and GATHER phases.
Class<IGASSchedulerImpl>	getSchedulerClass()
protected Callable<Long>	newFrontierStrategy(VertexTaskFactory<Long> taskFactory, IStaticFrontier f) Factory for the parallelism strategy that is used to map a task across the frontier.
<VS,ES,ST> IGASContext<VS,ES,ST>	newGASContext(IGraphAccessor graphAccessor, IGASProgram<VS,ES,ST> gasProgram) Obtain an execution context for the specified IGASProgram.
<VS,ES,ST> IGASState<VS,ES,ST>	newGASState(IGraphAccessor graphAccessor, IGASProgram<VS,ES,ST> gasProgram)
IGASSchedulerImpl	newScheduler()
IStaticFrontier	newStaticFrontier() Return an IStaticFrontier - this is the object that models the frontier that is consumed during a given round of evaluation.
void	setSchedulerClass(Class<IGASSchedulerImpl> newValue)
void	shutdown() Polite shutdown.
void	shutdownNow() Immediate shutdown.

Methods inherited from class java.lang.Object

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

Methods inherited from interface com.bigdata.rdf.graph.IGASEngine

getSortFrontier

Constructor Detail

GASEngine

public GASEngine(int nthreads)

Parameters:

indexManager - The index manager.

nthreads - The number of threads to use for the SCATTER and GATHER phases.

Method Detail

getNThreads

public int getNThreads()

Description copied from interface: IGASEngine

The parallelism for the SCATTER and GATHER phases.

Specified by:

getNThreads in interface IGASEngine

shutdown

public void shutdown()

Description copied from interface: IGASEngine

Polite shutdown.

Specified by:

shutdown in interface IGASEngine

shutdownNow

public void shutdownNow()

Description copied from interface: IGASEngine

Immediate shutdown.

Specified by:

shutdownNow in interface IGASEngine

newFrontierStrategy

protected Callable<Long> newFrontierStrategy(VertexTaskFactory<Long> taskFactory,
                                 IStaticFrontier f)

Factory for the parallelism strategy that is used to map a task across the frontier. The returned Callable should be executed in the caller's thread. The Callable will schedule tasks that consume the frontier. A variety of frontier strategies are implemented. Those that execute in parallel do so using the thread pool associated with the IGASEngine.

Parameters:

taskFactory - The task to be mapped across the frontier.

Returns:

The strategy that will map that task across the frontier.

getGASThreadPool

public ExecutorService getGASThreadPool()

If there is an ExecutorService for the GASEngine, then return it (nthreads GT 1).

Throws:

UnsupportedOperationException - if nthreads==1.

setSchedulerClass

public void setSchedulerClass(Class<IGASSchedulerImpl> newValue)

getSchedulerClass

public Class<IGASSchedulerImpl> getSchedulerClass()

newStaticFrontier

public IStaticFrontier newStaticFrontier()

Return an IStaticFrontier - this is the object that models the frontier that is consumed during a given round of evaluation. TODO Config overrides

newScheduler

public IGASSchedulerImpl newScheduler()

newGASContext

public <VS,ES,ST> IGASContext<VS,ES,ST> newGASContext(IGraphAccessor graphAccessor,
                                             IGASProgram<VS,ES,ST> gasProgram)

Description copied from interface: IGASEngine

Obtain an execution context for the specified IGASProgram.

Specified by:

newGASContext in interface IGASEngine

Type Parameters:

VS - The generic type for the per-vertex state. This is scoped to the computation of the IGASProgram.

ES - The generic type for the per-edge state. This is scoped to the computation of the IGASProgram.

ST - The generic type for the SUM. This is often directly related to the generic type for the per-edge state, but that is not always true. The SUM type is scoped to the GATHER + SUM operation (NOT the computation).

Parameters:

graphAccessor - Indicates the graph to be processed.

gasProgram - The program to execute against that graph.

newGASState

public <VS,ES,ST> IGASState<VS,ES,ST> newGASState(IGraphAccessor graphAccessor,
                                         IGASProgram<VS,ES,ST> gasProgram)