Package com.netflix.stats.distribution

This package contains a number of classes that produce statistics about on-the-fly accumulated data.

See:
Description

Interface Summary

DataCollector	An object that collects new values incrementally.
DataDistributionMBean	Abstract MBean interface for objects that hold information about a distribution of (double) values.
DistributionMBean	Abstract MBean interface for objects that describe the general characteristics of a distribution of (double) values.
HistogramMBean	Abstract MBean interface for objects that describe the general characteristics of a distribution of (double) values recorded as a histogram.

Class Summary

DataAccumulator	A double-buffer of DataBuffer objects.
DataBuffer	A fixed-size data collection buffer that holds a sliding window of the most recent values added.
DataDistribution	A DataAccumulator that also publishes statistics about the "previous" buffer.
DataPublisher	An object that periodically updates a DataAccumulator, swapping between the buffers.
Distribution	Accumulator of statistics about a distribution of observed values that are produced incrementally.
Histogram	This extends Distribution by tracking counts of values per "bucket" and the ability to find the (approximate) median value.

Package com.netflix.stats.distribution Description

This package contains a number of classes that produce statistics about on-the-fly accumulated data. Their primary use-case is collecting information about the distribution of API call response times. There are a few variations, with differing capabilities and memory and run-time costs. These classes all implement the DataCollector interface for adding new data, so that code that generates data need not know about the particular data collection implementation.

• The Distribution class is the lightest weight DataCollector. It reports basic information about the collected data, but gives no view into the shape of the distribution of that data. The reported information is updated and is available immediately.
• The Histogram class is slightly more expensive: It maintains counts of values in pre-defined ranges. This gives more information, but requires the user to have some idea of the proper ranges. Like the Distribution, reported information is updated and is available immediately.
• The class DataDistribution implements a double buffering scheme to support simultaneous accumulation of new data and analysis of the previous buffered data. This provides accurate percentile data without prior knowledge of the distribution of data, but requires more memory than the other collection types. Also, reports are generated when the buffers are swapped, and so are not immediately available. The class DataPublisher manages a background task that periodically swaps the data buffers and publishes the now "previous" data into a DataDistribution object.

These classes all implement MBean interfaces, and so can be viewed via JMX.