The price of abstraction

This article was featured on Slashdot today. It’s about the current state-of-the-art Linux-Desktop being quite demanding in Hardware – even more demanding than the arc-nemesis Windows XP.

And it’s true.

I see one of the problems in the basics of the Unix philosophy: Use small tools to do a specific task and another in the OpenSource-Philosophy: Write clean code.

These two approaches create wonderful architectures and abstractions of small tools doing their work.

What nobody seems to recognize: This so wonderful and well thought-out architecture is bloated per se. Let’s say you are playing a Video-File in a KDE-Video-Player running in KDE. This is what’s running on your system to acomplish this task (I hope I get all the (bigger) components really running – maybe there are more (or less) of them):

  • Linux Kernel
  • KDE-Sound-Server
  • X-Window-Server (complete with un-used network transparency which would not work with the video anyway
  • The whole QT-Library
  • Some basic KDE-Abstractions (kdelibs)
  • Your media player

Every one component ist cleanly seperated from each other – every one can be replaced without disturbing other components. Every one is designed cleanly using many abstractions to provide this replacability even for internal components.

But it get’s even more complicated: Many of the acting components are independant processes which creates the need for quite a bit of IPC wich is always slower than direct calls.

No wonder this is slow!

In Windows for example quite a lot of the stuff described above is actually running in the kernel or at least very close to it, maybe using undocumented interfaces to the kernel.

Playing a video mostly depends on DirectX which uses mostly in-process calls. It’s dirty, it’s unstable (maybe), but it’s fast, doesn’t flicker and happens to just work (the less independant components involved, the less can go wrong).

Of course that’s not how software should be written. It’s how it is written when fast and impressive resuslts are requested.