Include box_totals for http_1down/up

[eriktar]

I find the box_totals graph very useful in tcp_1up/down. Would be nice to have it along in the http tests as well.

[tohojo]

Do you mean a box plot of just the bandwidth, or one that includes HTTP fetch times?

[eriktar]

The one with bandwidth and http fetch time. If you look at the graphs produced at my side, the box_combine (which is in there pr default) is a bit hard to read. But the box is more readable.

PLOTS['box_combine'] = {'description': 'Box combination of HTTP latency and TCP bandwidth',

• 'type': 'box', # vs box_combine* 'axis_labels': ['Mean bandwidth (MBits/s)', 'Mean download time (ms)'], 'dual_axes': True, 'series': [ {'data': 'TCP download', 'label': 'TCP download', 'combine_mode': 'meta:MEAN_VALUE'}, {'data': 'HTTP latency', 'label': 'HTTP get (ms)', 'combine_mode': 'mean', 'axis': 2}]}

On Wed, Oct 17, 2018 at 9:01 PM Toke Høiland-Jørgensen < notifications@github.com> wrote:

Do you mean a box plot of just the bandwidth, or one that includes HTTP fetch times?

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/tohojo/flent/issues/150#issuecomment-430749681, or mute the thread https://github.com/notifications/unsubscribe-auth/AFGM4VMgaXKsFla5gm_I9xngYfh9Rwpdks5ul36jgaJpZM4XKJ_8 .

[eriktar]

Adding images into github as well.

[tohojo]

Yeah, the box_combine plot is not terribly useful with a single test. It plots the distribution of averages over several test runs; so try it with multiple files.

I'll add the other type as well :)

[eriktar]

Perhaps I'm doing it wrong, but I still find box more readable.

On Thu, Oct 18, 2018 at 10:16 AM Toke Høiland-Jørgensen notifications@github.com wrote:

Yeah, the box_combine plot is not terribly useful with a single test. It plots the distribution of averages over several test runs; so try it with multiple files.

I'll add the other type as well :)

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or mute the thread.

[tohojo]

eriktar notifications@github.com writes:

Perhaps I'm doing it wrong, but I still find box more readable.

Right, sure. Be aware that because box plots express percentiles, they can be misleading for data sets that do not consist of independent samples, such as timeseries data points from the same download.

I'm not saying they can't be useful; just be careful which conclusions you draw. For instance, saying things like "25% of downloads were below X Mbps" from a box plot would probably not be valid...