For the past few years I've been writing and speaking about
time-series storage, a problem inspired by what I help build
at VividCortex and also my personal
projects. I've been thinking about a related problem for much
longer, but I haven't had to work on it until now: ranking.
At VividCortex, we have a new tool called the Profiler. It allows
you to rank within various categories of metrics, like MySQL or
OS metrics, by a dimension, like count. You can rank top
processes by total CPU time or memory, MySQL queries by execution
time, and so on. We can do this for arbitrary time ranges.
The ranking is performed on the sums of values within
each time range. If we're ranking by CPU usage during a 300
second window, we look at the CPU usage measurements for each
process at every second, sum those values up, and …
[Read more]