Re: [LI] Message from RMS.

[Arun Sharma <adsharma@xxxxxxxxxxxxxxx>]

On Sat, Dec 04, 1999 at 04:02:48AM -0600, Manoj Srivastava wrote:
>         The HURD is several decades ahead of UNIX based Linux OS --
>  and cutting edge research operating systems take time to
>  stabilize. Linux, after all, is based on 30 year old technology. The
>  HURD is far newer.

(a) If you are implying that Hurd is better technology, because it is
    a microkernel, talk to Linus. He's known for his hatred of the 
    concept to the extent of showing up on comp.os.linux.advocacy on 
    a busy day when several top execs are trying to get him on phone :)

    He even accuses research people who were into microkernels as being
    either stupid or intellectually dishonest (to research micro kernels
    just to get funding).

    That's not to say - Linus is God and what he says must be the Bible
    attitude (I find this implication in popular press distasteful). My
    personal opinion is that micro kernels make sense on state of the art
    hardware where the cost of context switching is not high. But I'll
    trust Linus' judgement on what is good for Intel compatible hardware
    anyday, given his demonstrated ability to make good tradeoffs in
    simplicity vs performance. (Though I disagree with some of his
    choices)

(b) Linux is the most iconoclastic OS around, in ignoring a lot of 
    academic research in operating systems. CMU Mach has been the darling
    of academic research community for a whole decade. Yet, Linux hasn't
    incorporated anything from those research papers.

    In contrast, FreeBSD's virtual memory subsystem has been replaced with
    Mach's (ref: 4.4 BSD book by McKusick et al). BSDs are still very 
    strong in academic research in US universities (though people are moving
    to Linux due to its omnipresence). The paper on IOLite which received
    the best technical paper award at the last SOSP (Symposium on OS
    principles) was based on BSD.

    There is work going on to implement a pthreads implementation using
    scheduler activations (Tom Anderson's PhD thesis), see the first 
    reference in:

    http://docs.FreeBSD.org/cgi/getmsg.cgi?fetch=71688+0+archive/1999/freebsd-arch/19991114.freebsd-arch

    I can't see any such activity on linux-kernel.

	-Arun