[smokeping-users] Graph scale problem
tobi at oetiker.ch
Tue Apr 29 13:28:54 CEST 2008
yes you did find a number of corner cases here .. the problem realy
is what algorithm to use for the proper scaling. At the moment I
only look at the median and then add some space on top of it. But
maybe we should look at some other property ...
look for the find_max function in Smokeping.pm to see what is
presently being done ...
as for the logarithm stuff, this is a user contributed feature
which is not well integrated. I do all my work without it enabled,
I guess it would need rather different settings to work well.
For the upcomming ajax version of smoekping I imagine to have graph
properties configurable interactively so that you can make a graph
look right by hand once you have identified an interesting section.
Today G.W. Haywood wrote:
> Hello again Tobi,
> On Mon, 28 Apr 2008, G.W. Haywood wrote:
> > On Sat, 26 Apr 2008, Tobias Oetiker wrote:
> > > G.W. Haywood wrote:
> > >
> > > > http://www.jubileegroup.co.uk/JOS/misc/scales_oddity.gif
> > > >
> > > > Maybe graphs should default to a logarithmic scale?
> > >
> > > The reason the 'spike' is lower in the last 10 day graph is that a
> > > single pixle there represents more data than in the higher
> > > resolution graphs ... the net result is, that the spike becomes
> > > lower as it gets distributed across time ...
> > With the result that the graphs give a very misleading impression of
> > the data under certain circumstances.
> > ... you've made some changes in the latest version ... I'll have
> > a look at it before commenting further.
> The 'smoke' rising to the top of the graphs is an improvement. I'm
> now using logarithmic graphs but I'm still having difficulties with
> the scales.
> As you can see a minimum RTT on the logarithmic graphs of 100us gives
> rather silly results if the typical RTT is 30us...
> The range selection algorithm seems not to take (much?) into account
> deviations of RTT substantially above the norm. The results resemble
> a hedgehog after mowing the lawn. Possibly the graph parameters need
> to be individually configurable for each (probe,target) pair. FWIW my
> preference would probably be for manual configuration, at least of the
> graph range limits. My logic is that a particular network path will
> typically have a 'normal' performance. Large excursions from normal
> will be of most interest, while smaller fluctuations will tend to be
> ignored. I'm thinking that I would be quite happy with a fixed upper
> bound on the logarithmic graphs of ca. one second, and a lower bound
> which varies by factors of say 10e+0.5 from 10us to 100ms. Obviously
> anyone making ssh connections over a satellite link would want to say
> something about that. :) I'll tweak the code and see how it goes.
> Having griped more than enough about relatively easily fixed things
> like the graph scales I think I ought to thank you for all the hard
> work you've done to make SmokePing such a tremendously useful tool!
> smokeping-users mailing list
> smokeping-users at lists.oetiker.ch
Tobi Oetiker, OETIKER+PARTNER AG, Aarweg 15 CH-4600 Olten
http://it.oetiker.ch tobi at oetiker.ch ++41 62 213 9902
More information about the smokeping-users