Re: [Bloat] dslreports mockup

[Sebastian Moeller <moeller0@gmx.de>]

Hi jb,


On May 22, 2015, at 14:52 , jb <justin@dslr.net> wrote:

> Yes, that's a good idea. the library makes it awkward to mix data in one drill down but I can
> mark the idle benchmark somehow, perhaps a green background strip.

	That would work well for me ;)

> 
> It does show bands in the initial view, but they are not visible if none of the three bars
> extends high enough.

	Fair enough, I guess most important in the bar view is the direct comparison of the aggregate measures...

> Since they are averages, and if you get grade A or A+ they don't
> tend to extend high enough to show even the yellow band.
> 
> I guess I can see the next request: can the "uploading" and "download" un-exploded
> bars not be averages but be maximums.. or close to it … ?

	Bull’s eye ;) the war I see stats is that if one reports a mean value one also needs to report a variance measure, preferably a confidence limit (say 95% or even 99% ). But the RTT values (under load) decidedly are not normally distributed, so the mean might not be the best measure to begin with (maybe mean and skewness would work, the higher the skewness the worse the buffer bloat grade)...


Best regards
	Sebatian



> 
> On Fri, May 22, 2015 at 8:03 PM, Sebastian Moeller <moeller0@gmx.de> wrote:
> Hi jb,
> 
> this looks like a good way to allow easy comparison with the bars and detailed information about the time course, I like it.
> 
> On May 22, 2015, at 09:00 , jb <justin@dslr.net> wrote:
> 
> > Well the dual Y-Axis thing didn't work.
> > It would require removal of the color bands and looked confusing.
> >
> > So I've done a drill-down thing instead. You get just three bars, then can
> > drill into each by clicking, to see an expansion against its own Y-Axis.
> > Hard to explain, easier to see:
> >
> > http://www.dslreports.com/speedtest/525965
> 
>         I wonder, would it be possible to also show the last (few) idle data points from just before loading download and upload? That would allow to easily assess how much the latency jumps under load in the drilled down view. Also I think I would like to see the color bands in the bar plot as a first indication whether I need to drill into the detail-views at all ;)
> 
> Best Regards
>         Sebastian
> 
> >
> > On Fri, May 22, 2015 at 10:56 AM, Jonathan Morton <chromatix99@gmail.com> wrote:
> >
> > > On 22 May, 2015, at 03:17, jb <justin@dslr.net> wrote:
> > >
> > > Or I can just have two Y-Axis with auto-scaling on both.
> >
> > You could also try a square-root scale (as opposed to linear or logarithmic).  This should help with comparing data with different orders of magnitude, without flattening things as aggressively as a log scale.
> >
> > But perhaps we should see what it looks like before committing to it.
> >
> >  - Jonathan Morton
> >
> >
> > _______________________________________________
> > Bloat mailing list
> > Bloat@lists.bufferbloat.net
> > https://lists.bufferbloat.net/listinfo/bloat
> 
>