Re: bad interactive performance when building

[Johnny Billquist <bqt%softjar.se@localhost>]

Antti Kantee wrote:
> On Wed Feb 25 2009 at 19:42:57 +0100, Adam Hoka wrote:
> > Antti Kantee wrote:
> >
> > > Has anyone noticed that the UI slows down considerably on 5.0 when
> > > building with a high enough -j and doing something which consumes a
> > > little CPU for UI processes (e.g. minimize/maximize a web browser).
> > > This happens both with 4BSD and M2.  I never saw it on 4.99.42.
> > >
> > > I realize this may be simply because 5.0 is faster, but I don't know if
> > > we can call it progress if a faster system feels slower to the user.
> > Yeah I see this effect too. Running Xorg with nice -10 or even real time
> > priority usually makes a huge difference under load...
>
> That is obvious.  However, the question is why it wasn't necessary
> previously and how do we fix the default again so that users who do not
> necessarily know they need to run X with a negative nice (or builds with
> a positive one) continue to get good interactive performance.

Are we sure this isn't just a more visible effect of sluggish 
performance that some people have been complaining about for years. 
Others have just said that the people complaining just have a shortage 
of resources, and that the default tuning parameters can't fit all 
possible configurations...? Much fingers have been pointed at memory 
usage patterns penalizing interactive processes that happen to sit idle 
a lot, and thus are agressively swapped out when you have loads of other 
work going on that access lots of files.

        Johnny

--
Johnny Billquist                  || "I'm on a bus
                                  ||  on a psychedelic trip
email: bqt%softjar.se@localhost             ||  Reading murder books
pdp is alive!                     ||  tryin' to stay hip" - B. Idol