9
$\begingroup$

We're plotting time-series metrics in the context of network/server operations. The data has a 5-minute sample rate, and consists of things like CPU utilization, error rate, etc.

We're adding a horizontal "threshold" line to the graphs, to visually indicate a value threshold above which people should worry/take notice. For example, in the CPU utilization example, perhaps the "worry" threshold is 75%.

My team has some internal debate over what color this line should be:

  1. Something like a bright red that clearly stands out from the background grid and data lines, and indicates this is a warning condition
  2. Something more subtle and definitely NOT red, since the "ink" for the line doesn't represent any actual data, and thus attention shouldn't be drawn to it unnecessarily.

Would appreciate guidance / best practices...

$\endgroup$
10
$\begingroup$

If it does not break your styleguide I would rather color the background of the plots red/(yellow/)green than just plotting a line. In my imagination this should make it pretty clear to a user that values are fine on green and to be checked on red. Just my 5¢.

$\endgroup$
  • $\begingroup$ Totally agree. By making it a background it is clearly not data, but my making it coloured it shows a clear change in 'state'. $\endgroup$ – Ian Turner Jul 22 '10 at 9:04
  • 1
    $\begingroup$ If you don't want too much colour, just show a band of a neutral eg light grey - but do it in the foreground with some high level of transparency. This means that any part of the data line which shows above this threshold will automatically "pop" as it has greater contrast to the background than to your 'normal' band. $\endgroup$ – AdamV Jul 22 '10 at 17:11
  • $\begingroup$ Found a way to do this (high transparency overlay on the "danger" part of the graph) with our toolset, thanks! $\endgroup$ – MikeF Jul 26 '10 at 21:53
  • 2
    $\begingroup$ I agree that the background option is a good idea. But the red / green is not - my office mate is colour blind, and can't tell the difference. So even though it's good from the "traffic lights" point of view, it can cause problems! $\endgroup$ – Andrew May 16 '11 at 23:20
4
$\begingroup$

To me, whether or not the line represents actual data seems irrelevant. What's the point of the plot? If it's so that somebody will do something when utilization crosses a threshold, the line marking the threshold had better be very visible. If the point of the plot is to give an overview of utilization over time, then why include the line at all? Just put the major gridlines of your plot at intervals that will coincide with your threshold (25% in your example), and let the reader figure it out.

... y'all been reading too much Tufte.

$\endgroup$
  • 1
    $\begingroup$ Indeed some people are taking Tufte as gospel and not being particularly flexible... $\endgroup$ – MikeF Jul 21 '10 at 20:31
  • $\begingroup$ Leaving the reader to figure things out has been a source of countless problems. You have to communicate your message properly and if your message is the data over this line is an issue you must shown the line which is concerning. $\endgroup$ – Ian Turner Jul 22 '10 at 9:47
  • $\begingroup$ ... right. I wasn't suggesting he leave it to the reader if it was a part of his message. I was saying that if the threshold weren't a part of the plot, he could either leave it out entirely or incorporate a subtle visual cue to make the threshold easier to find if the reader wanted to look for it. $\endgroup$ – Matt Parker Jul 22 '10 at 16:09
3
$\begingroup$

If this is about your "Qnotifier" I think that you should plot the threshold line in some darker gray so it is distinguishable but not disturbing. Then I would color the part of the plot that reaches over the threshold in some alarmistic hue, like red.

$\endgroup$
  • $\begingroup$ No, nothing to do with Qnotifier, and that's not mine, it's a commercial product. The data I'm working with is very similar to that though. Completely agree with the "color the part over the threshold" but the plotting tools in use currently here don't allow that level of control. :-( $\endgroup$ – MikeF Jul 21 '10 at 20:42
  • $\begingroup$ Too bad... Still you can make the same trick only with the color of the threshold line. $\endgroup$ – user88 Jul 21 '10 at 21:23
  • $\begingroup$ @MikeF: Can your tool overplot? That is, plot a line with small dots at each data point, then plot (on top of that) larger circles only at the above-threshold points, which will result in a sort of bullseye at points above the threshold. Avoids the whole color-blind issue and leaves everything at the same contrast. $\endgroup$ – Wayne Aug 19 '11 at 18:10
  • $\begingroup$ Tool can't do that, but we break the data down into two series, one with just the points above threshold and one with the points below, and plot them on the same axis in different colors. Interesting... $\endgroup$ – MikeF Aug 20 '11 at 22:27
1
$\begingroup$

I would strongly enjoin you to avoid red as an indicator: there are many sorts of colour-deficiency that make this choice problematic (see eg http://en.wikipedia.org/wiki/Color_blindness#Design_implications_of_color_blindness ).

The high-contrast option is I believe the best choice.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.