[systemd-devel] [PATCH] sysctl.d: default to fq_codel, fight bufferbloat

[Lennart Poettering]

On Mon, 20.10.14 19:49, Thomas Meyer (thomas at m3y3r.de) wrote:

> 
> Am 20.10.2014 16:48 schrieb Lennart Poettering <lennart at poettering.net>:
> >
> > On Thu, 16.10.14 14:05, Michal Schmidt (mschmidt at redhat.com) wrote: 
> >
> > > Quoting from Jon Corbet's report of Stephen Hemminger's talk at Linux 
> > > Plumbers Conference 2014 (https://lwn.net/Articles/616241/): 
> > > 
> > >     [...] So Stephen encouraged everybody to run a command like: 
> > > 
> > >     sysctl -w net.core.default_qdisc=fq_codel 
> > > 
> > >     That will cause fq_codel to be used for all future connections (up 
> > >     to the next reboot). Unfortunately, the default queuing discipline 
> > >     cannot be changed, since it will certainly disturb some user's 
> > >     workload somewhere. 
> > > 
> > > So kernel developers know the best default is fq_codel, but won't 
> > > change to it in the kernel itself. Instead they expect distros to do it. 
> > > Let's have the recommended default in systemd. 
> >
> > Judging by the article this sounds like a good thing to do, and 
> > systemd seems to be an OK place for this change if the kernel folks 
> > don't want to make it. Hence, go ahead, but make sure to add this to 
> > NEWS with an explanation that downstream distros should be aware of 
> > the change, and possibly revert it, if they don't want to go down this 
> > route.
> 
> So what about:
> 
> kernel.kptr_restrict=1
> fs.protected_symlinks=1
> fs.protected_hardlinks=1

The latter two we already ship as defaults. 

The kptr thing, what's the story on that? Is there a clear opinion
that this is something to enable by default, but which isn't set from
upstream because of fear of compat probs?

Lennart

-- 
Lennart Poettering, Red Hat