[kwlug disc.] (advanced user only) Request for kernel compile time

[Andrew Kohlsmith (lists)]

On April 7, 2008 01:17:42 am Richard Weait wrote:
> make
> real    60m2.657s
> user    37m25.311s
> sys     5m19.111s

an hour to compile??

> 4-way Intel(R) Xeon(TM) CPU 3.00GHz stepping 04
>
> make
> real    44m23.687s
> user    40m47.005s
> sys     4m11.700s
>
> make -j5
> real    18m31.142s
> user    67m26.002s
> sys     5m18.785s

You've got to be kidding me; I have a crappy T2310 1.46GHz and I get under 5 
minute compile times.  A friend told me that it's not necessarily raw compute 
power but memory which makes a big difference.  I've got 4G on this laptop 
and under his careful tutelage run make -j50.  You have nothing to worry 
about unless you hit swap.

I believe his exact words were "If you hit swap, you die."  :-)

In fact, he says that you can run -j (with no number) and with 4G you should 
be able to fit the entire build in memory.  I tried that, and died. He later 
revealed that that wouldn't quite work when running X.  He's a sick pup.  :-)

Anyway; my suggestions based on my own testing are as follows:
1) As much RAM as possible.  Touching disk sucks.
2) Experiment with the number of simultaneous builds (-j ###); on my 
particular machine I found between -j 40 and -j 50 to be the best.
3) If you're really looking to improve build times, look at ccache.  I have 
also experimented with distcc, but for the kernel, ccache seemed to be the 
big help.  I was achieving roughly 2 minute full build times with it.

Now I realize that you're not looking to optimize your compile times; you're 
trying to see what kind of computing horsepower your system has.  Raw kernel 
compiles don't show you that.  Take that make -j50 and try it on a lessor 
machine... you'll start seeing the difference.  Without spawning a lot of 
processes, you're not taxing the processor as much as you can and thus not 
getting a good measure of your CPU's capabilities.  You need some application 
that is highly parallizable; make -j does that. :-)

-A.