Phoronix Test Suite 4.2-Randaberg Officially Released

[phoronix]

Phoronix: Phoronix Test Suite 4.2-Randaberg Officially Released

Version 4.2 of the Phoronix Test Suite open-source benchmarking software is now officially available. There's many new features to this quarterly update to this industry-leading cross-platform automated testing software...

http://www.phoronix.com/vr.php?view=MTI1Nzg

[mark45]

I wanted once to fix the color algorithm of the test suite but couldn't cause I don't know php.
Aside from the fact that php is mostly a language for the web and you don't have to worry about memory leaks, is there other reasons you're using php? If written in a native lang one could create actual executables for the OSes and I guess C/C++ are more appropriate for benchmarking and tapping into system properties.
I'm not lobbying for anything, I'm just wondering.

[Tgui]

Where did you get the idea for this release name? I lived in Randaberg Norway, just outside Stavanger, for a couple years!

[droste]

I wanted once to fix the color algorithm of the test suite but couldn't cause I don't know php.

If you have a good color choosing algorithm write it down in any language you want (well not an esoteric one please ;-) c, c++, python, java, javascript, perl, c#, vb, pascal, ruby, etc. are okay) and I can transfer it to php. I'm really interested in this, because I haven't found a good algorithm yet.

[mark45]

My first approach is (would be) to have a fixed array of like 10-15 different colors chosen by your eyeballs and make the test suite use them accordingly. For example, item1 from chart1 sticks to color1 across all charts from the benchmark.
I noticed a phoronix benchmark is almost always less than 5 items, top - like 10 items. In the unlikely event there's more than say 15 items - generate those with the current algorithm - this would be needed extremely rarely.

[droste]

This sounds like an easy way to do, but do you have 10-15 easy distinguishable colors (for all people and not only you)? I've seen good color palletes up to 6 colors, but if it gets to more than 6 colors there's at least two colors that some people can't easily distinguish.

[mark45]

Red, green, blue, black, brown - there, you have 10 easily distinguishable colors, ten because you take a dark and bright instance of each one (gray or light gray would be the bright instance of black), these colors alone cover like 99.9% of all phoronix benchmarks.

[droste]

Red, green, blue, black, brown - there, you have 10 easily distinguishable colors, ten because you take a dark and bright instance of each one (gray or light gray would be the bright instance of black), these colors alone cover like 99.9% of all phoronix benchmarks.

Again sounds good, but is only the theory. Do you have some color codes? (#000000, rgb(0, 0, 0) rgb(0%, 0%, 0%), hsl(0, 0%, 0%), cmyk(0%, 0%, 0%, 100%), ...)
My experience here is that as soon as the "light and dark" of some distinguishable colors come into play, it's not easily distinguishable anymore. For example red and brown in light and dark variants or dark variants of green, blue and black.

[mark45]

From what I've seen on phoronix, the crux of the matter is pretty much when 2 (very) close shades of the same color are chosen for 2 (or more) items in the same chart.

So it's not really about making sure the colors are very different (it's impossible), but about making sure the colors are different enough so that the average user can distinguish them with relatively little or no strain.

To help a little more, you can put the red, green, blue, brown and black colors first in the array, and then followed by their lighter shades, this way (since most benchmarks have up to 5 items) the difference in color will be very clear for like over 90% of the benchmarks from the get go with no effort.