[olug] Kernel Compilation

Vincent Raffensberger vraffensberger at home.com
Sat Dec 30 21:08:12 UTC 2000


I forgot something...  While you're compiling, switch to another console
and run "top -q".  From there you will see the number of gcc processes
working.  Without a -j option there should be only one.  When you
specify 4 there should be four.  On a dual processor system there will
be two on each processor and they will of course take twice as long to
finish as one, but no time will be lost starting the new gcc lines.

Vincent Raffensberger wrote:
> 
> No, on a dual processor machine 4 should be best.  This just specifies
> the number of gcc lines to process simultaneously.  I believe make waits
> for each one to complete before starting a new one.  When that is
> happening, idle cpu cycles are wasted waiting on disk access before
> starting the new line. <-???  Since we're referring to a real
> multi-tasking OS, this setting should help both single and smp systems,
> but I would raise it even higher with more than two processors.
> 
> (from the make man page)
> 
> -j jobs
>             Specifies  the  number  of jobs (commands) to run
> simultaneously.  If there is
>             more than one -j option, the last one is effective.  If the
> -j option is given
>             without  an  argument,  make  will  not  limit the number of
> jobs that can run
>             simultaneously.
> 
> Here's an experiment for you:  time one this way and then add another
> alias like this:
> 
> alias gcc='gcc -03'
> 
> and time that one.  That's a good P6 optimization.  There are a few
> others, but I think that one is the biggest.
> 
> Have you tried to optimize your disks with hdparm yet?  If you want to
> start tweaking things for performance, there are a lot of things you can
> do.  Many are simple changes to files in /proc and don't really require
> a kernel re-compile.
> 
> Jason Ferguson wrote:
> >
> > Ooooo, that was nice.  I slapped that one into my .bash_profile.  Gave me a
> > MASSIVE decrease in compilation time for the sane package. However, since I
> > have a 2 processor system, shouldnt I do 'make -j 2'?
> >
> > Jason
> >
> > Vincent Raffensberger wrote:
> >
> > > >From what I've seen, a little optimization for make gives a huge
> > > improvement (especially for smp).  Try this before you start compiling:
> > >
> > > alias make='make -j 4'
> > >
> > > I've used the "-03" option for egcs before, which should help, but I'm
> > > never in that much of a hurry.  "info egcs" will show you all the
> > > optimizations.  You could also recompile egcs, since it's probably just
> > > a i386 optimized binary from installation.
> > >
> > > Jason Ferguson wrote:
> > > >
> > > > Okay, yes, I do know how to compile a kernel, thank you very much. :)
> > > >
> > > > However, my question: is it possible or even advisable to use processor
> > > > optimizations when compiling a kernel?
> > > >
> > > > I ask this because, frankly, Im one of those geeks who likes to wring
> > > > the last ergs of performance out of my machine.  If I have a P3
> > > > processor, why shouldnt my stuff be optimizaed for it (even if I only
> > > > gain 10% or so?).
> > > >
> > > > What do you guys think?
> > > >
> > > > Jason
> > > >
> > > > ---------------------------------------------------------------------
> > > > To unsubscribe, e-mail: olug-unsubscribe at bstc.net
> > > > For additional commands, e-mail: olug-help at bstc.net
> > >
> > > ---------------------------------------------------------------------
> > > To unsubscribe, e-mail: olug-unsubscribe at bstc.net
> > > For additional commands, e-mail: olug-help at bstc.net
> >
> > ---------------------------------------------------------------------
> > To unsubscribe, e-mail: olug-unsubscribe at bstc.net
> > For additional commands, e-mail: olug-help at bstc.net
> 
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: olug-unsubscribe at bstc.net
> For additional commands, e-mail: olug-help at bstc.net

---------------------------------------------------------------------
To unsubscribe, e-mail: olug-unsubscribe at bstc.net
For additional commands, e-mail: olug-help at bstc.net



More information about the OLUG mailing list