org.jppf.server.scheduler.bundle.impl
Class AbstractAutoTunedBundler

java.lang.Object
  org.jppf.server.scheduler.bundle.AbstractBundler
      org.jppf.server.scheduler.bundle.impl.AbstractAutoTunedBundler

All Implemented Interfaces:

Bundler

Direct Known Subclasses:

AutoTunedBundler

public abstract class AbstractAutoTunedBundler
extends AbstractBundler

This class implements a self tuned bundle size algorithm. It starts using the bundle size defined in property file and starts changing it to find a better performance. The algorithm starts making The algorithm waits for some execution to get a mean execution time, and them make a change in bundle size Each time the change is done, it is done over a smaller range randomly selected (like Monte Carlo algorithm).

Author:

Domingos Creado

Field Summary

protected int	currentSize The currrent bundle size.
protected Random	rnd Used to compute a pseudo-random increment to the bundle size, as part of a Monte Carlo random walk towards a good solution.
protected Map<Integer,BundlePerformanceSample>	samplesMap A map of performance samples, aorted by increasing bundle size.

Fields inherited from class org.jppf.server.scheduler.bundle.AbstractBundler

bundlerNumber, profile, timestamp

Constructor Summary

AbstractAutoTunedBundler(AnnealingTuneProfile profile)
Creates a new instance with the initial size of bundle as the start size.

Method Summary

void	feedback(int bundleSize, double time) This method performs the actual bundle size computation, based on current and past performance data.
int	getBundleSize() Get the latest bundle size computed by this bundler.

Methods inherited from class org.jppf.server.scheduler.bundle.AbstractBundler

dispose, getBundlerNumber, getProfile, getTimestamp, maxSize, setup

Methods inherited from class java.lang.Object

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

Methods inherited from interface org.jppf.server.scheduler.bundle.Bundler

copy

Field Detail

currentSize

protected int currentSize

The currrent bundle size.

rnd

protected Random rnd

Used to compute a pseudo-random increment to the bundle size, as part of a Monte Carlo random walk towards a good solution.

samplesMap

protected Map<Integer,BundlePerformanceSample> samplesMap

A map of performance samples, aorted by increasing bundle size.

Constructor Detail

AbstractAutoTunedBundler

public AbstractAutoTunedBundler(AnnealingTuneProfile profile)

Creates a new instance with the initial size of bundle as the start size.

Parameters:

profile - the parameters of the auto-tuning algorithm, grouped as a performance analysis profile.

Method Detail

getBundleSize

public int getBundleSize()

Get the latest bundle size computed by this bundler.

Returns:

the bundle size as an int.

See Also:

Bundler.getBundleSize()

feedback

public void feedback(int bundleSize,
                     double time)

This method performs the actual bundle size computation, based on current and past performance data.
Depending on the the performance samples and profile parameters, the following actions may be triggered in this method:

• samples collection (unconditional)
• detection of performance profile changes, if not currently being done
• when a performance profile change is detected, recompute the bundle size.

Specified by:

feedback in interface Bundler

Overrides:

feedback in class AbstractBundler

Parameters:

bundleSize - bundle size of the new performance sample.

time - total execution time of the new sample.

See Also:

Bundler.feedback(int, double)