[lkml]   [1997]   [Nov]   [5]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectL4Linux is SLOWER than Linux
Please stop the bullshit about L4, MkLinux, and stuff!
it has been repeated that L4Linux was faster than Linux.
Well, this is plain false, and I invite you to read the paper
presented by Jochen Liedtke and his team at SOSP16 in last october (1997)
(Linus was there), that study in depth relative performance
of plain Linux, L4 Linux, and MkLinux.

"The Performance or µ-Kernel-Based Systems"
by Hermann Ha:rtig, Michale Hohmuth, Jochen Liedtke (now at IBM),
Sebastian Scho:nberg, and Jean Wolter

* L4 runs on i486, Pentium, and Alphas
* When running unchanged binaries, L4Linux is typically 5% *SLOWER*
than plain Linux, as compared to a noticeable 50% overhead for MkLinux.
* In worst case (piping data), L4 behaves is 13% slower than Linux,
whereas MkLinux is 3 times slower.
* Not recompiling the libc, and using the traditional syscall interface,
means yet more overhead for a trampoline.
* Similarly, putting MkLinux in user mode rather than colocating it
with the Mach kernel makes it even slower (L4Linux is in user mode).
* On the other hand, you can access the µK directly to achieve IPC
and user-level memory management, which can be significantly faster
than Linux IPC and mmap/mprotect() play. Also, real-time threads
and techniques such as cache partitioning are made possible with L4.

As a side note:
the principle of microkernel implementations is to add unnecessary
protection domain crossings, hence, slow down computers. The interesting
idea about the original microkernel design was to make things modular.
Linux is modular already (things could be even better, feel free to
contribute in that direction). Any "enhancement" made possible
by a microkernel could be made possible by hacking the linux kernel directly,
without having to do additional domain crossings.
[BTW, has anyone investigated stealing the optimized asm code from L4
to enhance Linux domain crossing?]

Follow-ups to /dev/null.


== Faré -=- (FR) François-René Rideau -=- (VN) Уng-Vû Bân -=- ==
Join a project for a free reflective computing system! | 6 rue Augustin Thierry
TUNES is a Useful, Not Expedient System. | 75019 PARIS FRANCE -=- Reflection&Cybernethics ==

 \ /
  Last update: 2005-03-22 13:40    [W:0.072 / U:3.440 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site