[distcc] How to involve the localhost too?

[Alexander.Farber at nokia.com]

Hi again,

> -----Original Message-----
> From: ext [mailto:Alexander.Farber at nokia.com]
> 
> > > However the tcc is only started on bolinux11, but not
> > > on the bolinux47 (ie. not on the localhost) - I don't
> > > see it with "top" and I don't see it in the logfile. This
> > > of course makes the build process slower, not faster.
> > Did you use make -j 8, or something like that?
> 
> No, I haven't. I've restarted it now with 
> 
> 	setenv DISTCC_HOSTS "bolinux11:1111 localhost"
> and 
> 	gmake -j4 CC="distcc tcc"
> 
> and see that tcc is now being called on the localhost too 
> and that distccd isn't needed on the localhost (right?) 
> 
> I'll report the build time improvement to you in 
> (hopefully ;-) few minutes

it took me a bit longer to test different cases
but here are the results. Which aren't very precise
because we have NFS Problems (and the builds below
have been performed on NFS partitions) and the PC's
aren't owned by me alone (there are other users and
processes running). But I'll try the builds on a
local partition and on separate PC's as soon as 
our Linux cluster here arrives...


THE "CREW":

1) bolinux47, Dell PC with 2 Xeons x 2.4 GHz, 2 GB RAM, 
   40GB ATA 100 IDE disk + 100 Mbit network card

2) bolinux11, IBM Intellistation, 1 Xeon x 1.7 GHz,
   512 MB RAM, SCSI Ultra 160 disk, 100 Mbit card.

3) bolinux54, IBM NetVista, 1 Xeon x 1.7 GHz

4) boccm03, Dell PC with 2 Xeons x 2.4 GHz, 2 GB RAM, 
   SCSI Ultra 160 disk + 100 Mbit network card


THE TESTS:

1) just bolinux47 (2 Xeons x 2.4 GHz, 2 GB RAM) alone

gmake -j4

2758.920u 530.290s 52:27.22 104.5%      0+0k 0+0io 5109641pf+0w
---------------------------------------------------------------
2) master: bolinux47 (2 Xeons x 2.4 GHz, 2 GB RAM)  
	 + bolinux11 (1 Xeon 1.7 GHz, 512 MB RAM)

gmake -j4 CC:="distcc tcc"

2364.220u 578.200s 1:04:12.25 76.3%     0+0k 0+0io 7328091pf+0w
---------------------------------------------------------------
3) master: bolinux47 (2 Xeons x 2.4 GHz, 2 GB RAM) 
	 + bolinux11 (1 Xeon 1.7 GHz, 512 MB RAM)
         + bolinux54 (1 Xeon 1.7 GHz, 256 MB RAM)

gmake -j8 CC:="distcc tcc"  	(zipped logfile attached)

2219.740u 616.770s 37:46.94 125.1%      0+0k 0+0io 5508460pf+0w
---------------------------------------------------------------
4) master: bolinux47 (2 Xeons x 2.4 GHz, 2 GB RAM) 
	 + bolinux11 (1 Xeon 1.7 GHz, 512 MB RAM)
         + bolinux54 (1 Xeon 1.7 GHz, 256 MB RAM)

gmake -j8 CC:="distcc tcc" C++:="distcc tcpp"

2254.010u 614.740s 38:25.19 124.4%      0+0k 0+0io 5526475pf+0w
---------------------------------------------------------------
5) master: bolinux47 (2 Xeons x 2.4 GHz, 2 GB RAM) 
	 + bolinux11 (1 Xeon 1.7 GHz, 512 MB RAM)
         + bolinux54 (1 Xeon 1.7 GHz, 256 MB RAM)
         + boccm03 (2 Xeons x 2.4 GHz, 2 GB RAM) 

gmake -j16 CC:="distcc tcc" C++:="distcc tcpp"

2196.920u 672.730s 29:41.13 161.1%      0+0k 0+0io 5528266pf+0w
----------------------------------------------------------------


NOTES:

The measurements aren't exact because there are other processes
and users on the machines. Also we have some NFS problems.

The 3) and 4) don't really differ, probably because there is 
not much C++ code to compile compared to amount of C code.

When executing 5) we've noticed, that only master is really
loaded - the other PC's are mostly idle and eventually 1 or 2
tcc processes show up in the "top" on those PC's, while you
see like 6 tcc processes in the "top" of bolinux47 (the master).


Also I have one QUESTION: would it be possible to utilize
distcc for generating *.d files, with something like:

	gmake -j8 CPP:="distcc gcc -E"

Regards
Alex


PS: One gotcha: first I used 

         gmake CC="distcc tcc" C++="distcc tcpp"

    but gmake hadn't really overriden the C++
    variable (it probably thought, I mean C+ +=). 
    So you have to use C++:="distcc tcpp" or 
    maybe: env "C++=distcc tcpp" gmake