Re: Bind ending up in Parked state. (pkg)

[Søren P. Skou <sps%t-rex.dk@localhost>]

Thus spake Anders Mundt Due, on Wed, Mar 11, 2015 at 11:49:33AM +0100:
> On 10 Mar, Søren P. Skou wrote:
> > > >is now happily chomping away, with a lot less load on the system overall.
> > > >Without --disable-threads it would ever so often spike all 8 Cores, and it
> > > >wasn't uncommon for it to have a load >2.0 - now it is sitting comfortably
> > > >around 0.32. And it doesn't seem to have any impact on the system.
> > > Still I would like to know what is taking all this time... ktrace it
> > > and then kdump -R to display relative timestamps.
> > anycast-dns-01# ldd /usr/pkg/sbin/named
> > /usr/pkg/sbin/named:
> >         -llwres.141 => /usr/pkg/lib/liblwres.so.141
> >         -lpthread.1 => /usr/lib/libpthread.so.1
> >
> > even with --disable-threads it still links into phtreads it seems?
> 
> Yeah, that was be having a bit of a fight with pkgsrc. Looks like I got
> this one figured.
Ah, that explained that one :) Anyhoo, since I had to put it back into
production it is now running without phtread (as you said it would ;-)
). It has been running for nearly 7 hours without a hitch now.

A top with t shows this for named:

12583     1 named     85 kqueue/6  55:26 13.77% 13.77% -         named


Load on the machine hasn't passed over 0.20 in load yet, whereas a named with
multithreading simply never got under 0.30 - Also, the machine seems a
lot more responsive.

As Andrea¿ Gustafsson said, he has been disabling threads on Named for
quite a while now. Shouldn't it be more efficient having more threads?
Provided of course that nothing actually blocks up :)

Best Regards 
S. P. Skou


-- 
GPG Key: 8E58ACB3