On Thu, Jun 08, 2006 at 08:34:28AM -0300, Gustavo Rios wrote:
> Dear friends,
> 
> i have been having an increasing interest on OpenBSD internals. So, my
> first movement was to buy some Unix books on OS design like The Design
> of the 4.4BSD  .....
> 
> One thing i would like to know better is about SMP. Up to now i could
> not figure it out about the algorithm choice and motivation, not i can
> have a good ideia about how openbsd does behave on an SMP system
> compared other unices like Solaris/AIX, etc or even pretenders like
> Linux.
> 
> Thank you a lot for your time and cooperation.

Unless I am sorely mistaken, OpenBSD uses the old - and slow - Big
Kernel Lock technique. This has the disadvantage that the kernel quickly
becomes a bottleneck; and for some applications, notably firewalling,
having more than one processor isn't likely to do any good (bsd.mp has
different interrupt handling though, and this seems to sometimes work
better than bsd's strategy; this is a different issue though, and I
don't know enough about it to say anything more sensible).

So, OpenBSD's SMP support isn't very good, performance-wise. On the
other hand, simple architectures have their own advantages: very few, if
any, bugs (compare to Linux, for instance!); no security problems; small
time requirements, so other parts can be improved; and it's easy to code
correctly.

I don't know about AIX, but Linux will be faster; FreeBSD will most
likely be faster, as well; and Solaris is quite possibly the best system
to be running on a 'very symmetric' system with lots of processors
and/or cores. At least performance-wise.

                Joachim

  • Re: SMP Joachim Schipper

Reply via email to