Re: Microbenchmarks

[Andrew Doran <ad%netbsd.org@localhost>]

On Thu, May 29, 2008 at 02:23:43PM +0200, 10.50%free.fr@localhost wrote:
> Selon Andrew Doran <ad%netbsd.org@localhost>:
> 
> > Hi,
> >
> > I thought I'd post this because it makes for interesting reading. It's the
> > output of Sun's libmicro, comparing these systems in (mostly) default
> > configurations: NetBSD-current, NetBSD 4.0, Fedora Core 9, FreeBSD 7.
> > The output and libmicro's home page:
> >
> >     http://www.netbsd.org/~ad/libmicro.html
> >     http://www.opensolaris.org/os/project/libmicro/
> >
> > Some tests are not included because they don't work properly on !Solaris
> > yet, couldn't complete because of resource limits (NetBSD 4) or crashed the
> > machine (not NetBSD). I made sure that the BSD systems were using the cpu's
> > cycle counter for clock, I think Linux already does that.
> 
> Could you be a bit more specific on this? Particulary what still needs
> improvement

Whatever bugs you :-). The numbers for fork and exec are quite prominent. I
wonder if something bad is happening in the pmap/uvm.

> and why we have got a so wonderful result on malloc (is it a
> property of the new jmalloc that it doesn't scale up CPU time in o(calls) but 
> in
> o(1) or so ?)

I don't know, but yes, it's basically because we switched to jemalloc.

Andrew