Sawtooth patterns (Group statistics are this evening's bugbear)

The problem with statistics is that it's often messy to calculate for measurements of the real world. You wind up needing to introduce all sorts of machinery either to artificially create a consistent measurement framework, or to track all the various exceptions and partial contributions so that the final measurement has some relation to what's actually happening.

For instance, the group-statistics mechanism in Cinemon is currently dropping the daily cummulative counters into the group summaries... problem is that those increase through the day, then drop at the end of the day, then increase again... the actual modems are just doing an approximate straight line. The daily summary data is needed for the modems (even daily summaries produce a *huge* swath of data to process), but the groups want to track and store the data for every single scan (since there are so few of them).


Comments are closed.


Pingbacks are closed.