Re: NetBSD in Set top box

[Jason R Thorpe <thorpej%zembu.com@localhost>]

On Wed, May 10, 2000 at 02:20:12PM +0200, anders.dinsen%lasat.com@localhost 
wrote:

 > and I think the MIT robot contest still uses the NetBSD kernel. In theory 
 > Linux
 > has one advantage over NetBSD when it comes to Java: that it supports kernel
 > threads (as cloned processes, but since Unix processes are almost as 
 > lightweight
 > as threads in other OS's, efficiency is fine). Kernel threads (along with a
 > threaded kernel) enables Java threads to take advantage of an SMP 
 > environment.

A couple corrections/clarifications:

        (1) The LinuxThreads model of one-process-per-thread (tho they
            all share address space, file descriptors, etc.) is actually
            well-known to be an inefficient threads model.  Linux processes
            are far from "lightweight", and even true LWPs as found on
            Solaris and, in a slightly different form, Tru64 UNIX (OSF/1)
            are still too heavy when you want to scale to many threads.

            Also, LinuxTreads is unable to implement some of the
            semantics required by POSIX as a side-effect of its
            fundamental nature... needless to say, I don't think
            very highly of LinuxThreads.

        (2) The Linux kernel is `threaded' only to the extent that the
            NetBSD kernel is.  That is, each process is a thread of
            execution in the kernel, and they run either to completion
            (return to userspace) or until they explicitly yield (e.g.
            tsleep() in the NetBSD kernel).  Neither kernel is fully
            preemptive (NetBSD-current does have a few special case
            preemption points, but it is far from a general solution).

-- 
        -- Jason R. Thorpe <thorpej%zembu.com@localhost>