Hi Ted, Hm, not sure if it's just me, but I don't see the graphs. I see their placeholders in the email, but they are not loading, and now I'm super curious to see, them, of course. :)
That said, I'm not sure if I'll be able to give you any precise feedback as we haven't measured QD very precisely, but once you open-source implementation we could run things side by side and provide feedback and maybe even blog about it. Otis -- Performance Monitoring * Log Analytics * Search Analytics Solr & Elasticsearch Support * http://sematext.com/ On Sat, Nov 16, 2013 at 1:31 PM, Ted Dunning <[email protected]> wrote: > > On Sat, Nov 16, 2013 at 10:10 AM, Otis Gospodnetic < > [email protected]> wrote: > >> We use it in SPM (performance monitoring). We've contributed a patch >> to lower the memory footprint for QDigest. The lib works well! :) >> > > > Hey Otis. If you look at these graphs, can you say anything about > size/accuracy trade-offs of QD relative to what I am building? > > These graphs tell you what size digest you need for different levels of > accuracy. If you only need 99%-ile and higher, you can use a 500Byte > digest and get 0.5% accuracy or better. With 2K, you get 1% for the > median, 0.5% for the 90-th %-ile and ~0.1% accuracy for 99% and higher. > > > > [image: Inline image 1] > > > [image: Inline image 3] >
