Class Algorithm
- java.lang.Object
-
- edu.isi.pegasus.planner.selector.site.heft.Algorithm
-
public class Algorithm extends java.lang.Object
The HEFT based site selector. The runtime for the job in seconds is picked from the pegasus profile key runtime in the transformation catalog for a transformation.The data communication costs between jobs if scheduled on different sites is assumed to be fixed. Later on if required, the ability to specify this value will be exposed via properties.
The number of processors in a site is picked by the attribute idle-nodes associated with the vanilla jobmanager for a site in the site catalog.
There are two important differences with the algorithm cited in the HEFT paper.
- Our implementation uses downward ranks instead of the upward ranks as mentioned in the paper. The formulas have been updated accordingly. - During the processor selection phase, we do the simple selection and not follow the insertion based approach.
- Version:
- $Revision$
- Author:
- Karan Vahi
- See Also:
AVERAGE_BANDWIDTH
,RUNTIME_PROFILE_KEY
,DEFAULT_NUMBER_OF_FREE_NODES
,AVERAGE_DATA_SIZE_BETWEEN_JOBS
,GridGateway.mIdleNodes
-
-
Field Summary
Fields Modifier and Type Field Description static float
AVERAGE_BANDWIDTH
The average bandwidth between the sites.static float
AVERAGE_DATA_SIZE_BETWEEN_JOBS
The average data that is transferred in between 2 jobs in the workflow.static int
DEFAULT_NUMBER_OF_FREE_NODES
The default number of nodes that are associated with a site if not found in the site catalog.private float
mAverageCommunicationCost
The average communication cost between nodes.static long
MAXIMUM_FINISH_TIME
The maximum finish time possible for a job.private java.lang.String
mLabel
The label of the workflow.private edu.isi.pegasus.common.logging.LogManager
mLogger
The handle to the LogManagerprivate edu.isi.pegasus.planner.common.PegasusProperties
mProps
The handle to the properties.private java.lang.String
mRequestID
The request id associated with the DAX.private java.util.Map
mSiteMap
Map containing the number of free nodes for each site.private java.util.List
mSites
The list of sites where the workflow can run.private edu.isi.pegasus.planner.catalog.site.classes.SiteStore
mSiteStore
Handle to the site catalog.private edu.isi.pegasus.planner.catalog.TransformationCatalog
mTCHandle
The handle to the transformation catalog.protected edu.isi.pegasus.planner.catalog.transformation.Mapper
mTCMapper
Handle to the TCMapper.private edu.isi.pegasus.planner.partitioner.graph.Graph
mWorkflow
The workflow in the graph format, that needs to be scheduled.static java.lang.String
PROCESS_CATALOG_IMPL_PROPERTY
The property that designates which Process catalog impl to pick up.static java.lang.String
RUNTIME_PROFILE_KEY
The pegasus profile key that gives us the expected runtime.
-
Constructor Summary
Constructors Constructor Description Algorithm(edu.isi.pegasus.planner.classes.PegasusBag bag)
The default constructor.
-
Method Summary
All Methods Instance Methods Concrete Methods Modifier and Type Method Description protected float
calculateAverageComputeTime(edu.isi.pegasus.planner.classes.Job job)
Returns the average compute time in seconds for a job.protected long[]
calculateEstimatedStartAndFinishTime(edu.isi.pegasus.planner.partitioner.graph.GraphNode node, java.lang.String site)
Estimates the start and finish time of a job on a site.protected float
computeDownwardRank(edu.isi.pegasus.planner.partitioner.graph.GraphNode node)
Computes the downward rank of a node.java.lang.String
description()
This method returns a String describing the site selection technique that is being implemented by the implementing class.protected long
getAvailableTime(java.lang.String site, long readyTime)
Returns the available time for a site.protected int
getExpectedRuntime(edu.isi.pegasus.planner.classes.Job job, edu.isi.pegasus.planner.catalog.transformation.TransformationCatalogEntry entry)
Return expected runtime.protected double
getExpectedRuntimeFromAC(edu.isi.pegasus.planner.classes.Job job, edu.isi.pegasus.planner.catalog.transformation.TransformationCatalogEntry entry)
Return expected runtime from the AC only if the process catalog is initialized.private float
getFloatValue(java.lang.Object key)
A convenience method to get the intValue for the object passed.protected int
getFreeNodesForSite(java.lang.String site)
Returns the freenodes for a site.long
getMakespan()
Returns the makespan of the scheduled workflow.java.lang.String
mapJob2ExecPool(edu.isi.pegasus.planner.classes.Job job, java.util.List pools)
The call out to the site selector to determine on what pool the job should be scheduled.protected void
populateSiteMap(java.util.List sites)
Populates the number of free nodes for each site, by querying the Site Catalog.void
schedule(edu.isi.pegasus.planner.classes.ADag dag, java.util.List sites)
Schedules the workflow using the heft.void
schedule(edu.isi.pegasus.planner.classes.ADag workflow, java.util.List sites, java.lang.String label)
Schedules the workflow according to the HEFT algorithm.protected void
scheduleJob(java.lang.String site, long start, long end)
Schedules a job to a site.
-
-
-
Field Detail
-
RUNTIME_PROFILE_KEY
public static final java.lang.String RUNTIME_PROFILE_KEY
The pegasus profile key that gives us the expected runtime.- See Also:
- Constant Field Values
-
PROCESS_CATALOG_IMPL_PROPERTY
public static final java.lang.String PROCESS_CATALOG_IMPL_PROPERTY
The property that designates which Process catalog impl to pick up.- See Also:
- Constant Field Values
-
AVERAGE_BANDWIDTH
public static final float AVERAGE_BANDWIDTH
The average bandwidth between the sites. In mega bytes/per second.- See Also:
- Constant Field Values
-
AVERAGE_DATA_SIZE_BETWEEN_JOBS
public static final float AVERAGE_DATA_SIZE_BETWEEN_JOBS
The average data that is transferred in between 2 jobs in the workflow. In megabytes.- See Also:
- Constant Field Values
-
DEFAULT_NUMBER_OF_FREE_NODES
public static final int DEFAULT_NUMBER_OF_FREE_NODES
The default number of nodes that are associated with a site if not found in the site catalog.- See Also:
- Constant Field Values
-
MAXIMUM_FINISH_TIME
public static final long MAXIMUM_FINISH_TIME
The maximum finish time possible for a job.- See Also:
- Constant Field Values
-
mAverageCommunicationCost
private float mAverageCommunicationCost
The average communication cost between nodes.
-
mWorkflow
private edu.isi.pegasus.planner.partitioner.graph.Graph mWorkflow
The workflow in the graph format, that needs to be scheduled.
-
mSiteStore
private edu.isi.pegasus.planner.catalog.site.classes.SiteStore mSiteStore
Handle to the site catalog.
-
mSites
private java.util.List mSites
The list of sites where the workflow can run.
-
mSiteMap
private java.util.Map mSiteMap
Map containing the number of free nodes for each site. The key is the site name, and value is aSite
object.
-
mTCMapper
protected edu.isi.pegasus.planner.catalog.transformation.Mapper mTCMapper
Handle to the TCMapper.
-
mLogger
private edu.isi.pegasus.common.logging.LogManager mLogger
The handle to the LogManager
-
mProps
private edu.isi.pegasus.planner.common.PegasusProperties mProps
The handle to the properties.
-
mRequestID
private java.lang.String mRequestID
The request id associated with the DAX.
-
mLabel
private java.lang.String mLabel
The label of the workflow.
-
mTCHandle
private edu.isi.pegasus.planner.catalog.TransformationCatalog mTCHandle
The handle to the transformation catalog.
-
-
Method Detail
-
schedule
public void schedule(edu.isi.pegasus.planner.classes.ADag dag, java.util.List sites)
Schedules the workflow using the heft.- Parameters:
dag
- theADag
object containing the abstract workflow that needs to be mapped.sites
- the list of candidate sites where the workflow can potentially execute.
-
schedule
public void schedule(edu.isi.pegasus.planner.classes.ADag workflow, java.util.List sites, java.lang.String label)
Schedules the workflow according to the HEFT algorithm.- Parameters:
workflow
- the workflow that has to be scheduled.sites
- the list of candidate sites where the workflow can potentially execute.label
- the label of the workflow
-
getMakespan
public long getMakespan()
Returns the makespan of the scheduled workflow. It is maximum of the actual finish times for the leaves of the scheduled workflow.- Returns:
- long the makespan of the workflow.
-
calculateEstimatedStartAndFinishTime
protected long[] calculateEstimatedStartAndFinishTime(edu.isi.pegasus.planner.partitioner.graph.GraphNode node, java.lang.String site)
Estimates the start and finish time of a job on a site.- Parameters:
node
- the node that is being scheduledsite
- the site for which the finish time is reqd.- Returns:
- long[0] the estimated start time. long[1] the estimated finish time.
-
computeDownwardRank
protected float computeDownwardRank(edu.isi.pegasus.planner.partitioner.graph.GraphNode node)
Computes the downward rank of a node.The downward rank of node i is _ ___ max { rank( n ) + w + c } j E pred( i ) d j j ji
- Parameters:
node
- theGraphNode
whose rank needs to be computed.- Returns:
- computed rank.
-
calculateAverageComputeTime
protected float calculateAverageComputeTime(edu.isi.pegasus.planner.classes.Job job)
Returns the average compute time in seconds for a job.- Parameters:
job
- the job whose average compute time is to be computed.- Returns:
- the weighted compute time in seconds.
-
getExpectedRuntime
protected int getExpectedRuntime(edu.isi.pegasus.planner.classes.Job job, edu.isi.pegasus.planner.catalog.transformation.TransformationCatalogEntry entry)
Return expected runtime.- Parameters:
job
- the job in the workflow.entry
- theTransformationCatalogEntry
object.- Returns:
- the runtime in seconds.
-
getExpectedRuntimeFromAC
protected double getExpectedRuntimeFromAC(edu.isi.pegasus.planner.classes.Job job, edu.isi.pegasus.planner.catalog.transformation.TransformationCatalogEntry entry)
Return expected runtime from the AC only if the process catalog is initialized. Since Pegasus 3.0 release it always returns -1.- Parameters:
job
- the job in the workflow.entry
- the TC entry- Returns:
- the runtime in seconds.
-
populateSiteMap
protected void populateSiteMap(java.util.List sites)
Populates the number of free nodes for each site, by querying the Site Catalog.- Parameters:
sites
- list of sites.
-
getFreeNodesForSite
protected int getFreeNodesForSite(java.lang.String site)
Returns the freenodes for a site.- Parameters:
site
- the site identifier.- Returns:
- number of nodes
-
scheduleJob
protected void scheduleJob(java.lang.String site, long start, long end)
Schedules a job to a site.- Parameters:
site
- the site at which to schedulestart
- the start time for jobend
- the end time of job
-
getAvailableTime
protected long getAvailableTime(java.lang.String site, long readyTime)
Returns the available time for a site.- Parameters:
site
- the site at which you want to schedule the job.readyTime
- the time at which all the data reqd by the job will arrive at site.- Returns:
- the available time of the site.
-
description
public java.lang.String description()
This method returns a String describing the site selection technique that is being implemented by the implementing class.- Returns:
- String
-
mapJob2ExecPool
public java.lang.String mapJob2ExecPool(edu.isi.pegasus.planner.classes.Job job, java.util.List pools)
The call out to the site selector to determine on what pool the job should be scheduled.- Parameters:
job
- Job theJob
object corresponding to the job whose execution pool we want to determine.pools
- the list ofString
objects representing the execution pools that can be used.- Returns:
- if the pool is found to which the job can be mapped, a string of the form
executionpool:jobmanager
where the jobmanager can be null. If the pool is not found, then set poolhandle to NONE. null - if some error occured .
-
getFloatValue
private float getFloatValue(java.lang.Object key)
A convenience method to get the intValue for the object passed.- Parameters:
key
- the key to be converted- Returns:
- the floatt value if object an integer, else -1
-
-