Le 24/11/2014 15:58, Pierre Labastie a écrit :
Hi,
Exceptionally top posting, because I try to answer (actually ask
questions) inside your text:
Le 24/11/2014 09:35, Christopher Gregory a écrit :
Hello,
I must say that even though they have improved the time to build this
version of java, that I am not in the least impressed that they did not
even bother to add the ability to specify parallel jobs to the build,
which would speed things up even more.
The docs say parallel jobs are enabled. I have not tested (but passed
--with-jobs=1 for SBU timings) that actually the parallel jobs are
enabled. What I saw is that the algorithm in "configure" is to take
the min of the installed memory in GB (rounded to an integer towards
zero), and the number of cores. For me, that algorithm leads to "2" as
a result (due to the amount of memory in the qemu VM I use). And I
think I saw a shorter build time (not kept the exact value, but around
7 SBUs), with this value. The docs say also that you can bypass that
calculation when using --with-jobs=N, which I have not tried, except
for N=1.
If they are going to modernize a build system you would think that they
would use someone who actually is experienced in writting Make files
that
support it.
This now brings me to the actual point of this email.
It still has listed in the Commands Explanations section:
--with-jobs=N
As this is not possible with this version, would it not be
appropriate to
comment this out, so that if they ever do learn in the future how to
write
proper Make files it can be easily re-instated again?
Well, at least it is useful to say that --with-jobs=1 works? Anyway,
doing what you say is easy, but I am a little reluctant. At least,
I'll try on bare metal (16GB memory, 8 cores), to see whether there is
any gain...
Yes I am being scathing about this as at least one of the developers
that
was working on iced tea works for the for PROFIT CORPORATION of Red Hat.
Thanks Pierre for updating to this so soon.
My sbu for this build was 15.6 as opposed to iced teas 29.5
Hum, cache memory dependant? I bought recently a new processor, which
has a bigger cache (I am at work now, and the said processor is at
home; will try to think of sending the exact value to night). My
timings oscillated between 9 and 10 SBUs (passing --with-jobs=1).
On my machine at work, letting the build system choose how many parallel
jobs to launch (configure found 14), here are the timings for
configure+make:
----------------------
Finished building OpenJDK for target 'all'
real 9m45.357s
user 55m28.092s
sys 2m22.024s
---------------------
which I understand as ~ 10 mn real time for ~ 1 h processor time. A gain
of almost 6x. Not too bad. On this machine, the SBU is ~165 s, so using
all the cores allow to be below 4 SBUs.
Pierre
--
http://lists.linuxfromscratch.org/listinfo/blfs-dev
FAQ: http://www.linuxfromscratch.org/blfs/faq.html
Unsubscribe: See the above information page
