Package org.acemq.workloads.metrics
Class LatencySummary
java.lang.Object
org.acemq.workloads.metrics.LatencySummary
Latency percentiles from one run.
The percentiles are reported and the mean is available but deliberately not printed first. A mean latency is the number least likely to describe anybody's experience: it hides the tail entirely, and the tail is what users notice and what capacity planning is about.
-
Method Details
-
empty
- Returns:
- an empty summary, for a measurement that recorded nothing
-
name
-
count
public long count()- Returns:
- how many observations this is built from
-
min
-
max
-
mean
-
p50
-
p90
-
p99
-
p999
- Returns:
- the 99.9th percentile: one request in a thousand is worse than this
-
p9999
-
percentile
- Parameters:
percentile- between 0 and 100- Returns:
- the value at it
-
tailRatio
public double tailRatio()- Returns:
- how many times worse the 99th percentile is than the median. Above about 10 the distribution has a tail worth explaining — a mean or a median alone would not show it
-
isEmpty
public boolean isEmpty()- Returns:
- whether anything was recorded
-
format
- Returns:
- a fixed-width line for a report
-
toString
-