2 ms·
To a Very crude first approximation, you could expect KDE to be 201 times slower than ratpoison, or rephrased you could have 201 times less user latency using r
by VLM 13y ago
To a Very crude first approximation, you could expect KDE to be 201 times slower than ratpoison, or rephrased you could have 201 times less user latency using ratpoison than KDE. Or ratpoison gets out of the way of the user "two hundred times faster" than KDE so for the average user it would be about that much more productive, with a correction factor for whatever proportion of time users play with WMs/desktops vs doing actual work in an app.
Also there's a fair number of machines out there, used for non-web browsing purposes, yet might still have a use for a GUI.
Lets say you want a virtual cloud host, connected to via VNC to ... linode. The smallest linode is a gig, so you'll be using 1/5 of your total resources just for the window manager if you use KDE. Is that relevant, well only you can decide.
Finally the most interesting part of the graph was the exponential-ish nature of it. For some weird reason I was guessing WMs/Desktops grow vaguely linearily over time so a random sample from random aged WMs would be vaguely linear when graphed... that's not whats happening. Have not come up with an adequate mental model to explain the shape of the curve yet, assuming the data gathering wasn't manipulated to specifically generate it. Maybe its an attractor effect where one pole pulls to 0 or as close as you can get, and the other pole pulls to infinity, or as close as you can get (kde) and everything smoothly distributes in the middle ground.
- jerf 13y ago"To a Very crude first approximation, you could expect KDE to be 201 times slower than ratpoison," Uselessly crude. What matters is how much code is on the critical path and how long it takes to run, not how much is loaded into RAM. A JPEG library loaded into memory that is not used when switching between windows doesn't impact window switching time. Contrariwise, KDE may have many thousands of times the latency if it tries to do many DBUS calls or something for an action that a simpler window manage would simply perform in-process. And it may not matter because "many thousands" of times the latency may still end up below the human perceptual threshold anyhow; 5ns vs 5ms is a whole lotta difference, except it sort of isn't when it comes to humans, both are instant.