[gentoo-user] Howto speed up compilations

[gentoo-user] Howto speed up compilations

[John J. Foster]

[-- Attachment #1: Type: text/plain, Size: 568 bytes --]

Good afternoon all,

A few weeks ago I read in one of the newgroups a way to greatly decrease
compilation times. The author noted that this was particularly noticable
when working with something like OO. The general jist of it was to
create temporary file system in memory and mount your portage tmpdir
there. For the life of me, I can't find that thread anymore. Does anyone
do something similar to this? Are there noticable gains to be had. I
have an Athlon 2800XP and 1 GB ram.

Thanks,
John

-- 
Contrary to the lie machine, the world is not safer.

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: [gentoo-user] Howto speed up compilations

[Bruno Lustosa]

On 7/18/05, John J. Foster <Gentoo-User@festus.150ml.com> wrote:
> A few weeks ago I read in one of the newgroups a way to greatly decrease
> compilation times. The author noted that this was particularly noticable
> when working with something like OO. The general jist of it was to
> create temporary file system in memory and mount your portage tmpdir
> there. For the life of me, I can't find that thread anymore. Does anyone
> do something similar to this? Are there noticable gains to be had. I
> have an Athlon 2800XP and 1 GB ram.

I am not sure if this will give a tremendous speedup. Granted, the
source files won't need to be read from disk, which is an advantage,
however, the file reading time should be very small compared to the
time it takes for the compiler to translate the source code into
machine code.
Also, there's the ammount of memory you will lose, memory that could
be used by the compiler. In some cases, gcc can eat very big chunks of
memory.
And if you use -pipe in your cflags, the gcc output isn't really
written to disk during the various stages of compilation. Instead,
it's piped through the processes.
Well... maybe someone will give you some accurate results.
Just my 2c.

-- 
Bruno Lustosa, aka Lofofora          | Email: bruno@lustosa.net
Network Administrator/Web Programmer | ICQ: 1406477
Rio de Janeiro - Brazil              |

-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] Howto speed up compilations

[Neil Bothwick]

[-- Attachment #1: Type: text/plain, Size: 1228 bytes --]

On Mon, 18 Jul 2005 14:14:41 -0300, Bruno Lustosa wrote:

> > A few weeks ago I read in one of the newgroups a way to greatly
> > decrease compilation times. The author noted that this was
> > particularly noticable when working with something like OO. The
> > general jist of it was to create temporary file system in memory and
> > mount your portage tmpdir there. For the life of me, I can't find
> > that thread anymore. Does anyone do something similar to this? Are
> > there noticable gains to be had. I have an Athlon 2800XP and 1 GB ram.
> 
> I am not sure if this will give a tremendous speedup. Granted, the
> source files won't need to be read from disk, which is an advantage,
> however, the file reading time should be very small compared to the
> time it takes for the compiler to translate the source code into
> machine code.
> Also, there's the ammount of memory you will lose, memory that could
> be used by the compiler. In some cases, gcc can eat very big chunks of
> memory.

Not to mention the OOo ebuild needing around 3GB of space in TMPDIR, so
this approach would only result in the emerge failing quicker.


-- 
Neil Bothwick

Capt'n! The spellchecker kinna take this abuse!

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: [gentoo-user] Howto speed up compilations

[Colin]

Neil Bothwick wrote:

>On Mon, 18 Jul 2005 14:14:41 -0300, Bruno Lustosa wrote:
>
>  
>
>>>A few weeks ago I read in one of the newgroups a way to greatly
>>>decrease compilation times. The author noted that this was
>>>particularly noticable when working with something like OO. The
>>>general jist of it was to create temporary file system in memory and
>>>mount your portage tmpdir there. For the life of me, I can't find
>>>that thread anymore. Does anyone do something similar to this? Are
>>>there noticable gains to be had. I have an Athlon 2800XP and 1 GB ram.
>>>      
>>>
>>I am not sure if this will give a tremendous speedup. Granted, the
>>source files won't need to be read from disk, which is an advantage,
>>however, the file reading time should be very small compared to the
>>time it takes for the compiler to translate the source code into
>>machine code.
>>Also, there's the ammount of memory you will lose, memory that could
>>be used by the compiler. In some cases, gcc can eat very big chunks of
>>memory.
>>    
>>
>
>Not to mention the OOo ebuild needing around 3GB of space in TMPDIR, so
>this approach would only result in the emerge failing quicker.
>  
>
Not if you've got a machine with more than 3 GB of memory.  A dual-proc 
Power Mac G5 can handle up to 8 GB of physical RAM.  If you did this 
trick on one of those, you might see some serious improvement!  But with 
most PC's being limited (by the x86 and motherboard designs) to 2 GB of 
physical RAM, it wouldn't work with large apps.

A good suggestion would be to grab some old computers, Gentoo-ize them, 
network them over 100BaseTX or Gigabit and make a little distcc farm.  
Plus, you can charge people if they want to come over and rent your 
computing power.  (Virginia Tech does that with their "System X," 1,100 
dual-2.3GHz-processor XServe G5's.)  :-)
--
Colin
-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] Howto speed up compilations

[John J. Foster]

[-- Attachment #1: Type: text/plain, Size: 420 bytes --]

On Mon, Jul 18, 2005 at 02:40:18PM -0400, Colin wrote:
> Plus, you can charge people if they want to come over and rent your 
> computing power.  (Virginia Tech does that with their "System X," 1,100 
> dual-2.3GHz-processor XServe G5's.)  :-)

As I'm a little short of this total right now, I think I'll stick with 
the -pipe option for now ;-)

John
-- 
Contrary to the lie machine, the world is not safer.

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: [gentoo-user] Howto speed up compilations

[Daniel da Veiga]

That sounded nice, if you can come with a "cluster" for compilation
you may see good results, I've never used distcc but a friend got
three athlon xps running and did it, he was able to install the three
gentoos within a day (wich is more than I could ever expect after my 2
days stage1 install with no x or openoffice).

On 7/18/05, Colin <signofzeta@gmail.com> wrote:
> Neil Bothwick wrote:
> 
> >On Mon, 18 Jul 2005 14:14:41 -0300, Bruno Lustosa wrote:
> >
> >
> >
> >>>A few weeks ago I read in one of the newgroups a way to greatly
> >>>decrease compilation times. The author noted that this was
> >>>particularly noticable when working with something like OO. The
> >>>general jist of it was to create temporary file system in memory and
> >>>mount your portage tmpdir there. For the life of me, I can't find
> >>>that thread anymore. Does anyone do something similar to this? Are
> >>>there noticable gains to be had. I have an Athlon 2800XP and 1 GB ram.
> >>>
> >>>
> >>I am not sure if this will give a tremendous speedup. Granted, the
> >>source files won't need to be read from disk, which is an advantage,
> >>however, the file reading time should be very small compared to the
> >>time it takes for the compiler to translate the source code into
> >>machine code.
> >>Also, there's the ammount of memory you will lose, memory that could
> >>be used by the compiler. In some cases, gcc can eat very big chunks of
> >>memory.
> >>
> >>
> >
> >Not to mention the OOo ebuild needing around 3GB of space in TMPDIR, so
> >this approach would only result in the emerge failing quicker.
> >
> >
> Not if you've got a machine with more than 3 GB of memory.  A dual-proc
> Power Mac G5 can handle up to 8 GB of physical RAM.  If you did this
> trick on one of those, you might see some serious improvement!  But with
> most PC's being limited (by the x86 and motherboard designs) to 2 GB of
> physical RAM, it wouldn't work with large apps.
> 
> A good suggestion would be to grab some old computers, Gentoo-ize them,
> network them over 100BaseTX or Gigabit and make a little distcc farm.
> Plus, you can charge people if they want to come over and rent your
> computing power.  (Virginia Tech does that with their "System X," 1,100
> dual-2.3GHz-processor XServe G5's.)  :-)
> --
> Colin
> --
> gentoo-user@gentoo.org mailing list
> 
> 


-- 
Daniel da Veiga
Computer Operator - RS - Brazil

-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] Howto speed up compilations

[John J. Foster]

[-- Attachment #1: Type: text/plain, Size: 1059 bytes --]

On Mon, Jul 18, 2005 at 01:07:38PM -0400, John J. Foster wrote:
> Good afternoon all,
> 
> A few weeks ago I read in one of the newgroups a way to greatly decrease
> compilation times. The author noted that this was particularly noticable
> when working with something like OO. The general jist of it was to
> create temporary file system in memory and mount your portage tmpdir
> there. For the life of me, I can't find that thread anymore. Does anyone
> do something similar to this? Are there noticable gains to be had. I
> have an Athlon 2800XP and 1 GB ram.
> 
> Thanks,
> John

OK, this happens all the time. I search, can't find what I want, post a
question, search again, and there it is. This is not the "thread" I was
talking about, but it was right in front of me on the Gentoo Wiki.
http://gentoo-wiki.com/TIP_Speeding_up_portage_with_tmpfs

Does anyone have any "tips" on these "tips"?

John
> 
> -- 
> Contrary to the lie machine, the world is not safer.



-- 
Contrary to the lie machine, the world is not safer.

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: [gentoo-user] Howto speed up compilations

[Zac Medico]

John J. Foster wrote:
> 
> OK, this happens all the time. I search, can't find what I want, post a
> question, search again, and there it is. This is not the "thread" I was
> talking about, but it was right in front of me on the Gentoo Wiki.
> http://gentoo-wiki.com/TIP_Speeding_up_portage_with_tmpfs
> 
> Does anyone have any "tips" on these "tips"?
> 

I sincerely doubt that this technique is worth the trouble because the Linux virtual memory system will automatically cache frequently accessed files in ram.  The tmpfs will eliminate filesystem overhead and will certainly result in a reduction of total build time.  However, I'd go with Bruno's recommendation and just put -pipe in the cflags.  For builds that work well in parallel, which most large builds do, distcc is a great way to speed things up.

Zac
-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] Howto speed up compilations

[Richard Fish]

Zac Medico wrote:

> John J. Foster wrote:
>
>>
>> OK, this happens all the time. I search, can't find what I want, post a
>> question, search again, and there it is. This is not the "thread" I was
>> talking about, but it was right in front of me on the Gentoo Wiki.
>> http://gentoo-wiki.com/TIP_Speeding_up_portage_with_tmpfs
>>
>> Does anyone have any "tips" on these "tips"?
>>
>
> I sincerely doubt that this technique is worth the trouble because the 
> Linux virtual memory system will automatically cache frequently 
> accessed files in ram.  The tmpfs will eliminate filesystem overhead 
> and will certainly result in a reduction of total build time.  
> However, I'd go with Bruno's recommendation and just put -pipe in the 
> cflags.  For builds that work well in parallel, which most large 
> builds do, distcc is a great way to speed things up.
>
> Zac


I'm a bit dubious about this well...unless you have a huge amount of 
memory (as in, more than 2G).  But even then, if you have sufficient 
RAM, most of the source files should still be in memory as a result of 
the extraction of the source archive before compilation.  And if buffers 
get flushed/recycled as a result of the compilation, that generally 
means that the compiler needed a large amount of memory, and giving the 
compiler your RAM is a much better choice, as otherwise swap has to be 
used, and that will *kill* your compilation time!

That isn't to say that I don't think some smart and simple things beyond 
-pipe can speed up builds.  I believe /var should be it's own 
filesystem, about 5G in size, and positioned just before /usr if you 
have one, or just after / if not.  The filesystem should be either XFS 
(caches files in memory very agressively, and in fact doesn't even 
update the disk for short-lived temporary files) or reiserfs v3 (creates 
and deletes files very very quickly).

However, I can pretty much guarantee that nothing short of a 
16-processor, 16GB system (or a distcc farm, as Colin suggested) will 
allow you to "emerge -Dv --emptytree kde-meta" in 15 minutes!!!  There 
is simply no getting around the CPU and memory bandwidth required for 
compiling!

-Richard

-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] Howto speed up compilations

[David Morgan]

On 13:07 Mon 18 Jul     , John J. Foster wrote:
> Good afternoon all,
> 
> A few weeks ago I read in one of the newgroups a way to greatly decrease
> compilation times. The author noted that this was particularly noticable
> when working with something like OO. The general jist of it was to
> create temporary file system in memory and mount your portage tmpdir
> there. For the life of me, I can't find that thread anymore. Does anyone
> do something similar to this? Are there noticable gains to be had. I
> have an Athlon 2800XP and 1 GB ram.
> 
> Thanks,
> John
> 

There was a thread about ways of speeding up compilation on gentoo-dev
(including what you mentioned) a while ago.

Good look compiling OOo with $PORTAGE_TMPDIR mounted in ram though ;)

Dave


-- 
gentoo-user@gentoo.org mailing list

[gentoo-user] Re: Howto speed up compilations

[Francesco Talamona]

On Monday 18 July 2005 19:07, John J. Foster wrote:
> Good afternoon all,
>
> A few weeks ago I read in one of the newgroups a way to greatly
> decrease compilation times. The author noted that this was
> particularly noticable when working with something like OO. The
> general jist of it was to create temporary file system in memory and
> mount your portage tmpdir there. For the life of me, I can't find
> that thread anymore. Does anyone do something similar to this? Are
> there noticable gains to be had. I have an Athlon 2800XP and 1 GB
> ram.

IIRC OO takes more than 3GB of space to compile. Gcc itself is quite RAM 
demanding...

I tkink it is feasible for smaller programs.

Ciao
	Francesco
-- 
Linux Version 2.6.11-gentoo-r11, Compiled #1 Thu Jun 23 05:26:18 
One 2.2GHz AMD Athlon 64 Processor, 2GB RAM, 4308.99 Bogomips Total
aemaeth
-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] Re: Howto speed up compilations

[Nick Rout]

On Mon, 18 Jul 2005 19:26:38 +0200
Francesco Talamona wrote:

> On Monday 18 July 2005 19:07, John J. Foster wrote:
> > Good afternoon all,
> >
> > A few weeks ago I read in one of the newgroups a way to greatly
> > decrease compilation times. The author noted that this was
> > particularly noticable when working with something like OO. The
> > general jist of it was to create temporary file system in memory and
> > mount your portage tmpdir there. For the life of me, I can't find
> > that thread anymore. Does anyone do something similar to this? Are
> > there noticable gains to be had. I have an Athlon 2800XP and 1 GB
> > ram.
> 
> IIRC OO takes more than 3GB of space to compile. Gcc itself is quite RAM 
> demanding...
> 
> I tkink it is feasible for smaller programs.

yeah and smaller programs take a negligible time in most cases.

anyone who has tested this, did you have ccache turned off? if it is on
it will skew results to hell, as a lot of the object files will be
cached.


> 
> Ciao
> 	Francesco
> -- 
> Linux Version 2.6.11-gentoo-r11, Compiled #1 Thu Jun 23 05:26:18 
> One 2.2GHz AMD Athlon 64 Processor, 2GB RAM, 4308.99 Bogomips Total
> aemaeth
> -- 
> gentoo-user@gentoo.org mailing list

-- 
Nick Rout

-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] Howto speed up compilations

[Rafer]

On Monday 18 July 2005 19:07, John J. Foster wrote:
> Good afternoon all,
>
> A few weeks ago I read in one of the newgroups a way to greatly decrease
> compilation times. The author noted that this was particularly noticable
> when working with something like OO. The general jist of it was to
> create temporary file system in memory and mount your portage tmpdir
> there. For the life of me, I can't find that thread anymore. Does anyone
> do something similar to this? Are there noticable gains to be had. I
> have an Athlon 2800XP and 1 GB ram.
>
> Thanks,
> John

Yes Sure :)  Here John read This Sektion ->

 http://gentoo-wiki.com/TIP_Speeding_up_portage
-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] Howto speed up compilations

[Zac Medico]

Rafer@gmx.de wrote:
>  http://gentoo-wiki.com/TIP_Speeding_up_portage

That's a portage optimization rather than a compilation optimization.  If you're doing portage optimization then you might also want to put /var/cache/edb/dep on a special filesystem.  In fact, portage supports depcache plugins for further optimization.

Zac
-- 
gentoo-user@gentoo.org mailing list