On Thu, 6 Jul 2000, Jeffry Smith <[EMAIL PROTECTED]> wrote:
...
>
> This isn't totally a linux problem, but the distributions need to pay
> more for people to write documentation. Of course, the redone
> linuxdoc.org and Open Writers project are helping.
I agree, and it is getting better all the time. I just wanted to try to
tease apart the aspects of Paul's claim "most comprehensive documentation
of any OS available"
> > > Scalability
> > > -----------
>
> Not true. What you're referring to was the Mindcraft tests, which
> caught the fact that (a year ago), the TCP/IP stack on Linux was
> single-threaded.
Whew, I thought you were indicating I was *advocating* the Mind(-less)
craft tests were valid! (I would never say that. It was rigged)
I do agree with you that those tests did sting us in that they showed
the relatively poor SMP I/O performance of Linux kernel at the time.
> Thus, using multiple NICS, and serving static pages,
> MS NT (which had just gotten a threaded TCP/IP stack, with the latest
> service pack at the time), could supersaturate a T3 line, vs merely
> saturating it. C't's tests of dynamic content, and using a single
> GigaBit Ethernet card, showed that Linux could beat MS in those cases.
> However, the community took the tests to heart, and (at least in the
> 2.3 series, but I beleive the 2.2 series as well), Linux is now using
> a multi-threaded TCP/IP stack.
Yes, but how many processors and NICS are we talking about here? Just
4 (or maybe 8), or larger numbers? My guess is both Linux and NT poop
out about 8 (even on better non-Intel h/w), but I haven't followed this
and so don't know what the benchmarks currently are.
However, I feel having Linux scale well past 8 processors is really not
high priority now (some macho kernel guys may think otherwise), since
that is really a rare space. Who really cares now that Linux currently
wouldn't scale well on, say, Sun's 64 and 112 processor monsters? I'm
positive Linux will always adapt more than quickly enough to hardware
that becomes more commonly available.
That said, I can't say I know the path Linux would take to become a
fully multi-threaded kernel like Solaris, since that would likely
require a big redesign...
Karl Runge
**********************************************************
To unsubscribe from this list, send mail to
[EMAIL PROTECTED] with the following text in the
*body* (*not* the subject line) of the letter:
unsubscribe gnhlug
**********************************************************
