Re: [PLUG] BSD Preferences?

[gabriel rosenkoetter <gr@eclipsed.net>]

On Mon, May 20, 2002 at 02:59:28PM -0400, Adam Turoff wrote:
> Yep.  AFAICT, the only way to get a BSD running on a cheap 2-processor 
> celeron box is to run FreeBSD.

Or do a cvs -d anoncvs.netbsd.org checkout -r SMP src (or something
like that; not sure of the exact tag, since I have no multi-processor
machines). But you get what you asked for. (One doesn't branch code
one thinks is ready for prime time. ;^>)

> OpenBSD *might* be SMP at the moment,

Nope, they won't steal that till it gets into the trunk. ;^>

> FreeBSD has worked on dual-CPU mobos for a 
> couple of years now (since 3.x, IIRC).  It's functional but I'm not
> enough of a hardware geek to know what they're doing right/wrong;
> FreeBSD 5.0 is supposed to have some killer improvements with SMP.

Last I heard they were still using a Big Lock (as in, if any thread
locks kernel data structures, no thread can get at *any* of the data
structures till the first thread is done). It's possible that the
improvements in 5.0 are fine-grained locking.

(I should really be keeping up with that. Multi-processors in a
single machine don't interest me much, but clustering on top of
NetBSD does, and some of the principles are the same.)

> Also, SMP on *BSD sorta tops out at 2 or possibly 4 CPUs; for real
> SMP hardware with an open OS, you're best off with a recent Linux
> distro.  (Or a proprietary OS.)

Eh? There's no good theoretical reason for such a limit, but I'm
willing to believe it doesn't work right for stupid reasons.

If you actually want multi-threaded stuff that hits the kernel hard,
though, you want a Sun Enterprise server and Solaris 8 (well, 9,
really, but they won't give it to you yet). If you can't afford
that, you don't actually need multi-threaded stuff that hits the
kernel hard, and pthreads on a single, fast processor will probably
do the job just fine. :^> > > Z.  > > >

-- 
gabriel rosenkoetter
gr@eclipsed.net

Attachment: pgpfRDZGRvb7I.pgp
Description: PGP signature