Code originally written 2006-2008, documented on my old blog
Inspiration from my 2006 CMG/2007 HPTS paper and slide deck "Utilization is Virtually Useless as a Metric", so what should we use instead? Bottom line: "Utilization is properly defined as busy time as a proportion of elapsed time. The replacement for utilization is headroom which is defined as the unused proportion of the maximum useful throughput."
To use:
> source("chp.r")
Definition and options
chp <- function(throughput,response, q=0.95, qx=F, xl="Throughput",yl="Response",tl="Throughput Over Time", ml="Headroom Plot", fit=T, max=T, splits=0)
throughput, response: array of values, both must be the same length
q=0.95 : default 95%ile outlier trimming for response time values
qx=F : default false don't also trim throughput outliers
xl="Throughput" : default X-axis label
yl="Response" : default Y-axix label
tl="Throughput..." : default time series label
ml="Headroom Plot" : default main plot label
fit=T : show 1/x fit curve by default
max=T : show maximum response staircase by default
splits=0 : set this to split the timeseries data into multiple colored sets
To test:
> chp(1:10,1:10)
This will open a plot window showing the following image. It shows a scatterplot of response time as a function of throughput, percentile outliers can be removed from the response time or both metrics. It shows a histogram of throughput distribution oriented outside the X-axis, response time distribution oriented outside the Y-axis, and a small view of sequential throughput over time for the data. The scatterplot is optionally annotated with a staircase showing the maximum value of the response time in each histogram bin, and an attempt is made to fit a throughput weighted inverse (1/x) curve to the data to find the "knee in the curve" for where the response time starts to increase rapidly with throughput.
Also see the slides from my 5-minute ignite talk on Bottleneck Analysis
Github repo created to celebrate #TLAPD2016 Arrrrrrr, R
