Trending the data over time is better left to other tools that trend data over time. The memcached stats should be a snapshot in time that other tools can consume. If you start trending things in memcached you not only waste resources but you will end up with metrics like unix load average that are meaningless because they can't be directly compared to anything else.
On Thu, Apr 9, 2009 at 9:42 AM, Dustin <dsalli...@gmail.com> wrote: > > > On Apr 8, 8:54 pm, Eric Lambert <eric.d.lamb...@gmail.com> wrote: > > Fair point .... so this raises a question about who is the audience for > > the stats. If we expect the stats to be mainly consumed by tools and > > monitoring frameworks, then I agree that we should just generate raw > > numbers, but if we expect people to be directly using the stats, then I > > think there is some value in these calculated numbers, even though they > > are less than precise. Should we split the difference and generate both > > the raw number and the calculated ratio? > > As much as I hate the phrase, it sounds like ``just do both'' makes > perfect sense here. > > 1) Find out how much memory is in use, and how much is actually > required *right now*. > 2) Trend the data over time. > > Two variables increment -- only one decrements. > > (sounds better if you say it with your dramatic voice) -- Eric Bergen eric.ber...@provenscaling.com http://www.provenscaling.com
