I’d like to point out that the definition of a DTU is that it’s “a blended measure of CPU, memory, and data I/O and transaction log I/O…” None of the perfmon counters used by the DTU Calculator take memory into account, but it is clearly listed in the definition as being part of the calculation. This isn’t necessarily a problem, but it is evidence that the DTU Calculator isn’t going to be perfect.
I’ll upload some synthetic load into the DTU Calculator, and see if I can figure out how that black box works. In fact, I’ll fabricate the CSVs completely so that I can totally control the perfmon numbers that we load into the DTU Calculator. Let’s step through one metric at a time. For each metric, we’ll upload 25 minutes (1500 seconds–I like round numbers) worth of fabricated data, and see how that perfmon data is converted to DTUs.
Andy then goes on to show how the DTU Calculator estimates DTU usage given different resource patterns. It’s a very interesting process and Andy clarified it considerably.