Resample shortens a long series so a graph stays readable: the evaluated period is cut into equal time windows, and the points of each window become one point, dated at the start of the window.

It is set in the Reduce section of the editor:

• Resample target points: the number of windows, 60 by default
• Disable resample: every point is kept

Resample is the last step of a DPS, after Reduce. It only applies to a time series:

Reduce scope	Resample
(none)	each raw series
Merged series	the merged series, see Merged series below
Single value, One per series	no effect: the result is already one value per series

Resample never adds points: a series with no more points than Resample target points is returned as it is, and a window without any point is left out (the External API keeps it, empty).

How a window becomes one point

The window value follows the metric type and, with VALUE, the kind of the metric:

Metric type	Kind	Window value
VALUE	average (response time, CPU %)	Σ(value × count) / Σcount
VALUE	interval counter (Responses Per Interval)	sum
VALUE	fluctuating counter (Concurrent Invocations)	highest
VALUE	other	average
MAX	any	highest
MIN	any	lowest
COUNT	any	sum: the number of values recorded in the window

The new point carries the sum of the counts of its window.

• The kind is not chosen in the DPS: it is read from the metric (examples per kind in Metric semantics).
• An average without count (points from an older Enterprise Manager extension) gives a plain average.

The reduce function is never used: the reduce combines several series at one instant, resample combines one series over time.

Example, a Merged series with Sum of CPU %, resampled into 15 min windows:

Window 10:00 → 10:15	Value
60 points of total CPU (Merged series, Sum), one every 15 s	about 45 % each
Window by kind (average)	45 %: the total CPU over the window
Window with the reduce function Sum	2,700 %: 60 instants added up

Example

• Resample target points = 2 over 10:00:00 → 10:01:00: two windows of 30 s
• a point is dated at the end of the 15 s it covers: 10:00:30 goes to window 1, which ends there

	Window 1 10:00:00 → 10:00:30		Window 2 10:00:30 → 10:01:00
Point	10:00:15	10:00:30	10:00:45	10:01:00
Value	1	2	3	4
Count	1	3	1	1

Metric type, kind	Window 1, dated 10:00:00	Window 2, dated 10:00:30
VALUE, average	(1×1 + 2×3) / 4 = 1.75	(3×1 + 4×1) / 2 = 3.5
VALUE, interval counter	1 + 2 = 3	3 + 4 = 7
VALUE, fluctuating counter	2	4
VALUE, other	(1 + 2) / 2 = 1.5	(3 + 4) / 2 = 3.5
MAX / MIN	2 / 1	4 / 3
COUNT (the values are the counts)	1 + 3 = 4	1 + 1 = 2
Count carried by the new point	4	2

A point before the period goes to the first window, a point after it to the last one.

Merged series

A Merged series can merge metrics of different kinds: its kind has to be determined, to pick its row in the resample table. The reduce function sets that kind, one for the whole merged series, and the count of each of its points.

Reduce function	Merged value	Merged count	Kind used by resample
Sum	Σ value	highest count	kind of the series, other if their kinds differ
Average	average value	average count (a typical series)	kind of the series, other if their kinds differ
Average weighted	Σ(value × count) / Σcount	Σcount	kind of the series, other if their kinds differ
Sum weighted	Σ(value × count)	Σcount	interval counter: a total adds up over time
Max / Min	value of the selected series	count of the selected series	kind of the series, other if their kinds differ
Count series	number of series	highest count	other: a number of series is averaged

Examples, with metric type VALUE:

Merged series	Its kind	Row of the resample table
Sum of CPU % of 2 JVMs	average	Σ(value × count) / Σcount
Sum of Responses Per Interval	interval counter	sum
Sum of Average Response Time + Responses Per Interval	other: the kinds differ	average
Count series of any metric	other	average
Sum weighted of any metric	interval counter	sum

The merged count only weighs the point with metric type VALUE on an average.