tech-kern archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

re: high sys time, very very slow builds on new 24-core system

> On Fri, 25 Mar 2011, Andrew Doran wrote:
> > On Fri, Mar 25, 2011 at 12:16:32AM -0700, Erik Fair wrote:
> > 
> > > It will be interesting to see if this class of problem shows up again when
> > > NetBSD is supported on the UltraSPARC "Niagra" CPUs.
> > 
> > Agree.  Although as it stands 24 cores/threads would be a problem for
> > sparc64 since it has a single mutex around the pmap.  One of the goals of
> > the uvmplock branch is to make it very easy to address that problem.
> The pmap needs to be rewritten for sun4v anyway since you can't access the 
> hardware directly any more.

i dunno about rewritten, but it needs some clean ups.

FWIW, i've already made a bunch of removals of pmap_lock for sparc64
on the rmind-uvmplock branch (uncommited.)  the clean up andy is
talking about is pretty minor.  there's more to go yet, and i haven't
really benchmarked this at all yet, but the uvmplock branch definately
seems more stable for the sparc* platforms than HEAD right now for SMP.

it is however, not quite finished.


Home | Main Index | Thread Index | Old Index