Nice to read if you ever wondered why our Histogram is so different and what is that MixMaxCounter we talk about: http://kamon.io/core/metrics/instruments/ …
@giltene @kamonteam in that case the HdrHistogram would tell you very little beyond the mean and the fact of the very low variance.
-
-
@avibryant@kamonteam Yup. If you configure it with 3 decimal point accuracy, all numbers equal to within that will look the same. -
@avibryant@kamonteam But that's not an "interesting" problem area for measuring latency percentiles… -
@giltene@kamonteam yep. I'm coming from a machine learning context, where being able to eg find that "DC offset" on the fly is more useful. -
@avibryant@kamonteam In this case it's trivial at the start: E.g. take the average of the first 100 results in the stream before feeding. -
@giltene@kamonteam bit trickier in a distributed context; each node will have a slightly different average, then you have to reconcile. -
@avibryant@kamonteam Reconciling isn't hard. Each node can produce their own histogram, and they are additive later. -
@giltene@kamonteam right, but if they each have a different offset it gets harder. -
@giltene@kamonteam this isn't an academic question - I'm thinking in particular about using it in MapReduce, where the nodes have no comm. - 4 more replies
New conversation -
-
-
@giltene@kamonteam but I suspect these cases are extremely rare (and if you stumble onto one, a simple linear transformation will fix it). -
@avibryant@kamonteam Yeh, if you have such data, you can just establish and subtract the "DC offset" level before you feed HdrHistogram.
End of conversation
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.