Performance data classification

Performance Monitoring Infrastructure provides server-side data collection and client-side API to retrieve performance data. Performance data has two components: static and dynamic.

The static component consists of a name, ID and other descriptive attributes to identify the data. The dynamic component contains information that changes over time, such as the current value of a counter and the time stamp associated with that value.

The PMI data can be one of the following statistical types defined in the JSR-077 specification...

• CountStatistic

• BoundaryStatistic

• RangeStatistic

• TimeStatistic

• BoundedRangeStatistic

RangeStatistic data contains current value, as well as lowWaterMark and highWaterMark.

In general, CountStatistic data require a low monitoring level and TimeStatistic data require a medium monitoring level. RangeStatistic and BoundedRangeStatistic require a high monitoring level.

There are a few counters that are exceptions to this rule. The average method response time, the total method calls, and active methods counters require a high monitoring level. The Java Virtual Machine Profiler Interface (JVMPI) counters, SerializableSessObjSize, and data tracked for each individual method method(level data) require a maximum monitoring level.

In previous versions, PMI data was classified with the following types:

• Numeric: Maps to CountStatistic in the JSR-077 specification. Holds a single numeric value that can either be a long or a double. This data type is used to keep track of simple numeric data, such as counts.

• Stat: Holds statistical data on a sample space, including the number of elements in the sample set, their sum, and sum of squares. You can obtain the mean, variance, and standard deviation of the mean from this data.

• Load: Maps to the RangeStatistic or BoundedRangeStatistic, based on JSR-077 specification. This data type keeps track of a level as a function of time, including the current level, the time that level was reached, and the integral of that level over time. From this data, you can obtain the time-weighted average of that level. For example, this data type is used in the number of active threads and the number of waiters in a queue.

These PMI data types continue to be supported through the PMI API. Statistical data types are supported through both the PMI API and Java Management Extension (JMX) API.

The TimeStatistic type keeps tracking many counter samples and then returns the total, count and average of the samples. An example of this is an average method response time. Given the nature of this statistic type, it is also used to track non-time related counters, like average read and write size. You can always call getUnit method on the data configuration information to learn the unit for the counter.

In order to reduce the monitoring overhead, numeric and stat data are not synchronized. Since these data track the total and average, the extra accuracy is generally not worth the performance cost. Load data is very sensitive, therefore, load counters are always synchronized. In addition, when the monitoring level of a module is set to max, all numeric data are also synchronized to guarantee accurate values.

 

See Also

Monitoring performance

 

---