Subject: RE: Aberrant behaviour -- any experience?
From: Stanislav Sinyagin (ssinyagin@yahoo.com)
Date: Tue May 21 2002 - 16:27:16 CDT
--- Jake Brutlag <jakeb@microsoft.com> wrote: > The algorithm can only "converge" to the degree that the underlying data > is consistent. But you can probably get some better results by tuning > the algorithm. I would suggest (1) increasing the intercept adaption > (alpha), seasonal adaptation (gamma), and/or seasonal deviation > (gamma-deviation) parameter, try 0.2 or 0.25; this will help the > algorithm adjust to the steep inclines you see Sat/Sun/Mon and/or Thanks, I'll try to play with the parameters tomorrow. > (2) > changing the failures threshold/window length to reduce the number of > failures in general (although we use the default). I already reduced it to 2 failures in 30-minute window. > You can use the > rrdtool tune command to change these on existing files. After making the > changes, you can force reinitialization to occur by using the > aberrant-reset option. As far as I understood from the manual, aberrant-reset option does not re-apply the algorithm to the existing data, and I can see the effect only few days later? Why not implementing a new option, aberrant-reapply, which would re-scan the existing data and re-calculate all holt-winters data accordingly? For me it would be easier to make a perl script which clones the RDD and does this, but I think it's much more useful to have it in RRDtools themselves. I'll try and find some time to look into the sources, maybe it's not so difficult to do it myself... Thanks, Stan __________________________________________________ Do You Yahoo!? LAUNCH - Your Yahoo! Music Experience http://launch.yahoo.com -- Help mailto:majordomo@net.doit.wisc.edu and say "help" in message body Unsubscribe mailto:majordomo@net.doit.wisc.edu and say "unsubscribe flowscan" in message body Archive http://net.doit.wisc.edu/~plonka/list/flowscan/archive/
This archive was generated by hypermail 2b25 : Tue May 21 2002 - 16:29:12 CDT