Compute faban metrics/statistics

[Cerfoglg]

The faban data we save from the transformers already contain most of the trial level metrics and statistics we want. The only one that needs computing come from the detail.xan files, which need statistics to be computed for both the WebDriver Throughput and WebDriver Response Times.

At experiment level, for every data we have from faban, we need to compute our statistics.

From our summary.xml files we then have to compute from data:

• totalOps (normal avg)
• operation successes
• operation failures
• operation mix (metrics for each single operation, as well as for the sum of all operations)
• operation response times stats
• operation delay times average (weighted by the totalOps avg)
• operation delay times min
• operation delay times max

From the custom stats, we could either need to compute our statistics or metrics however. Because these are dynamic, we need to know what needs to be done with those custom stats.

The boolean values have to be aggregated using an AND at experiment level.

@VincenzoFerme We need a way to know what needs to be done on the custom stats. Either we can insert a new field in the xml to tell us, or insert a tag in the description field.

[VincenzoFerme]

@Cerfoglg the actual possible entity of a CustomMetric element in the Faban data are described on the following link: http://faban.org/1.3/docs/api/com/sun/faban/driver/CustomMetrics.Element.html. @simonedavico we should support all of them.

Don't assume they are all there, but if they are, aggregate them as done for the similar field in other Faban metrics.

We can use the target to distinguish statistic from absolute numbers.

[VincenzoFerme]

We postpone for now the support of the aforementioned reference to the Custom Metrics support in https://github.com/benchflow/drivers-maker and related services. The support is ready in the analysis part of the BenchFlow framework.

[simonedavico]

In the future, in order to support this feature, a spoon Processor should be implemented for each kind of CustomMetric; the processor would know what fields to add to the drivers classes, and what methods to call in the operation methods.

Of course, CustomMetric implementations should be added accordingly.

[VincenzoFerme]

Evaluate why faban, reports less request that number of models we actually have in the database. Maybe it is due the fact that Faban probably collect data only during the steady state. If that is the case, we might want to evaluate to consider the Faban steady state to select the data on the server side.