Re: Laptop exhibits erratic responsiveness

On Mon, Nov 30, 2020 at 12:48:47PM -0800, David Wolfskill wrote:
> On Mon, Nov 30, 2020 at 05:21:28AM -0800, David Wolfskill wrote:
> > ...
> > FWIW, the copy of the laptop's "head" kernel config in
> > http://www.catwhisker.org/~david/FreeBSD/head/CANARY is current.
> > 
> > And the only difference between the stable/12 config and the head config
> > is:
> > 
> > g1-48(12.2-S)[3] diff -u FreeBSD-1{2,3}/CANARY
> > --- FreeBSD-12/CANARY   2019-07-08 10:43:51.940026000 -0700
> > +++ FreeBSD-13/CANARY   2015-03-22 10:03:41.000000000 -0700
> > _at__at_ -342,7 +342,7 _at__at_
> >  # from some parts of the kernel.  As this makes everything more noisy,
> >  # it is disabled by default.
> >  #
> > -# options      DIAGNOSTIC
> > +options        DIAGNOSTIC
> >  
> >  #
> >  # REGRESSION causes optional kernel interfaces necessary only for regression
> > 
> > Hmmm... so I suppose that re-disbling DIAGNOSTIC might be a reasonable
> > thing to try, as well, yeah...?  :-}
> > ....
> 
> Well, success(ish):
> 
> * I was able to reproduce the observed effect using the newer laptop,
> running a kernel from 14 Nov (thus, r367674):
> 
> * I was able to reproduce the observed effect using the newer laptop,
>   running a kernel from 14 Nov (thus, r367674).  (I had cloned the old
>   laptop's disk drive and slapped it inito the newer laptop.)
> 
> * I then updated the kernel config file ("CANARY") to comment out the
>   DIAGNOSTIC option, then rebuilt/installed the kernel.
> 
> * Upon reboot, "ping albert" showed remarkably consistent RTTs (compared
>   to before) -- e.g., the initial example I posted showed:
> 
>   --- albert.catwhisker.org ping statistics ---
>   67 packets transmitted, 67 packets received, 0.0% packet loss
>   round-trip min/avg/max/stddev =3D 0.428/1024.248/68482.244/8303.513 ms
> 
>   On the newer laptop after the above change, I saw:
> 
>   --- albert.catwhisker.org ping statistics ---
>   30 packets transmitted, 30 packets received, 0.0% packet loss
>   round-trip min/avg/max/stddev = 0.463/0.574/0.671/0.054 ms
> 
> Please especially note the maxima & standard deviations.
> 
> 
> FWIW, I do have (in /boot/loader.conf):
> 
> # Try to reduce WITNESS overhead
> vm.debug.divisor=4
> 
> (which I also utilize at work).
> 
> I suppose I will be commenting out that DIAGNOSTIC option now.... :-}

Try to just set debug.vmem_check to 0 and see if it helps.