Re: packages compressed with xz

2010-12-05 Thread Lars Engels
On Sat, Dec 04, 2010 at 11:19:21PM +0100, Simon L. B. Nielsen wrote:
 
 On 30 Nov 2010, at 03:16, jhell wrote:
 
  Agreed. Soon can be quantified by actual need and of which there is not
  much need except for larger packages but adding this would just add
  unneeded complication to the system that is already in place.
 
 
 We are running out of diskspace on event the FTP master site -
 currently we are at ~1TB. The xz compression gives as significant
 space saving - so there is already a need.
 
 PS. anyone saying a 1 TB etc. disk is cheap will be ignored.

And all ftp mirror admins will be very grateful to sync smaller
.xz packages. :)


pgpADHhbR7zTC.pgp
Description: PGP signature


Re: packages compressed with xz

2010-12-05 Thread Ion-Mihai Tetcu
On Fri, 03 Dec 2010 23:07:51 +0200
Volodymyr Kostyrko c.kw...@gmail.com wrote:

 30.11.2010 04:40, Julien Laffaye wrote:
  You can specify limits during compression, so the question is
  should we do that so that hosts with N MB of RAM can decompress
  packages?  Do we retain the compression ratio over bzip2 if we
  limit compression memory to 512 MB so that decompression would be
  possible with, say, 128 MB?
 
  According to xz(1), in its default mode (-6), xz uses ~100MiB for
  compression and ~10MiB for decompression.
  That seems to be acceptable.
 
 You possibly miss something about compression/decompression.
 
 The designated memory size is not directly affected only by
 compression mode. When decompressing you will need memory for:
 
 1. Data history.
 2. Dictionary.
 3. Some indexes.
 
 And those ones are all empty at start. So say, if you are compressing 
 something really huge trying to use 4G of memory you end using that
 much memory between 2G - 3G of source data. And we will need 512MB to 
 decompress that hunk of data.
 
 Are the packages _that_ large?
 [ .. ]

The biggest package that can be produced by a port it's a bit over 10G.

-- 
IOnut - Un^d^dregistered ;) FreeBSD user
  Intellectual Property is   nowhere near as valuable   as Intellect
FreeBSD committer - ite...@freebsd.org, PGP Key ID 057E9F8B493A297B


signature.asc
Description: PGP signature


Re: packages compressed with xz

2010-12-05 Thread Eitan Adler
 The biggest package that can be produced by a port it's a bit over 10G.


Curious - which one?


-- 
Eitan Adler
___
freebsd-ports@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-ports
To unsubscribe, send any mail to freebsd-ports-unsubscr...@freebsd.org


Re: packages compressed with xz

2010-12-05 Thread Robert Huff

Eitan Adler writes:

   The biggest package that can be produced by a port it's a bit over 10G.
  
  Curious - which one?

OpenOffice(-3)?


Robert Huff

___
freebsd-ports@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-ports
To unsubscribe, send any mail to freebsd-ports-unsubscr...@freebsd.org


Re: packages compressed with xz

2010-12-05 Thread Ion-Mihai Tetcu
On Sun, 5 Dec 2010 11:41:29 -0500
Eitan Adler li...@eitanadler.com wrote:

 
   Curious - which one?
 
         OpenOffice(-3)?
 
 He told me on IRC - something in games/
 
 $grep -R NO_PACKAGE /usr/ports/games :-)

Hum, looks like either they got smaller or QAT doesn't have all the
games packages built yet:
-rw-r--r--   1 root  wheel   5.3G Nov 29 02:42 
crafty-tablebase-pawn-20070910.tbz
-rw-r--r--   1 root  wheel   1.7G Nov 29 01:59 
crafty-tablebases-no-pawn-20070910.tbz

-- 
IOnut - Un^d^dregistered ;) FreeBSD user
  Intellectual Property is   nowhere near as valuable   as Intellect
FreeBSD committer - ite...@freebsd.org, PGP Key ID 057E9F8B493A297B


signature.asc
Description: PGP signature


Re: packages compressed with xz

2010-12-05 Thread Volodymyr Kostyrko

05.12.2010 18:03, Ion-Mihai Tetcu wrote:

And those ones are all empty at start. So say, if you are compressing
something really huge trying to use 4G of memory you end using that
much memory between 2G - 3G of source data. And we will need 512MB to
decompress that hunk of data.

Are the packages _that_ large?

  [ .. ]

The biggest package that can be produced by a port it's a bit over 10G.



0k, I'll try to be sharper. How many of that huge packages will run on a 
prehistoric cpu/64Mb mem combo?


I'll bet even working with the OpenOffice will be... challenging.

Let's take the other route. Packages that large will benefit from 
maximum xz compression most. The question clearly is Would we like to 
ditch maximum compression for those huge ones taking care of the 
hardware that would never be ready to run them?


Don't listen to me anyway. Our first goal is clearly making FreeBSD work 
everywhere. Capping the xz limits would do nothing for me personally 
because all of my machines are capable of building any software I want - 
I don't use packages at all. So I don't have a right to vote. We should 
think from the point of people running that hardware. And I think we 
should support them.


--
Sphinx of black quartz judge my vow.

___
freebsd-ports@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-ports
To unsubscribe, send any mail to freebsd-ports-unsubscr...@freebsd.org


Re: packages compressed with xz

2010-12-04 Thread Simon L. B. Nielsen

On 30 Nov 2010, at 03:16, jhell wrote:

 Agreed. Soon can be quantified by actual need and of which there is not
 much need except for larger packages but adding this would just add
 unneeded complication to the system that is already in place.


We are running out of diskspace on event the FTP master site - currently we are 
at ~1TB. The xz compression gives as significant space saving - so there is 
already a need.

PS. anyone saying a 1 TB etc. disk is cheap will be ignored.

-- 
Simon L. B. Nielsen

___
freebsd-ports@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-ports
To unsubscribe, send any mail to freebsd-ports-unsubscr...@freebsd.org


Re: packages compressed with xz

2010-12-03 Thread Volodymyr Kostyrko

30.11.2010 04:40, Julien Laffaye wrote:

You can specify limits during compression, so the question is should we do that
so that hosts with N MB of RAM can decompress packages?  Do we retain the
compression ratio over bzip2 if we limit compression memory to 512 MB so that
decompression would be possible with, say, 128 MB?

According to xz(1), in its default mode (-6), xz uses ~100MiB for
compression and ~10MiB for decompression.
That seems to be acceptable.


You possibly miss something about compression/decompression.

The designated memory size is not directly affected only by compression 
mode. When decompressing you will need memory for:


1. Data history.
2. Dictionary.
3. Some indexes.

And those ones are all empty at start. So say, if you are compressing 
something really huge trying to use 4G of memory you end using that much 
memory between 2G - 3G of source data. And we will need 512MB to 
decompress that hunk of data.


Are the packages _that_ large?

I think that in worst possible case we need to double the size of 
package for comfort decompression. If the lower point is 64Mb then 32Mb 
package compressed with any compression strategy would be decompressed 
without hitting a swap.


We'll need someone to do actual testing but look at this one:

# ls -la ImageMagick-6.6.5-10.tar.xz
-rw-r--r--  1 root  wheel  6316324 21 ноя 23:52 ImageMagick-6.6.5-10.tar.xz

# time xz -dt ImageMagick-6.6.5-10.tar.xz
0.860u 0.036s 0:01.02 87.2% 68+1492k 56+0io 1pf+0w

--
Sphinx of black quartz judge my vow.

___
freebsd-ports@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-ports
To unsubscribe, send any mail to freebsd-ports-unsubscr...@freebsd.org


Re: packages compressed with xz

2010-11-30 Thread perryh
Ion-Mihai Tetcu ite...@freebsd.org wrote:

  It would be nice to support xz(1) compression for large selective
  packages like firefox or openoffice as those will never run on
  smaller systems.

 Trouble is it ain't no way (CPU, space, banhdwidth on our side and
 space,bandwidth on our mirrors side) we could build a double set of
 packages.

but perhaps it would be possible to support compressing certain
ports' packages using xz _instead of_ bzip2, the specification
being made in the port's Makefile and defaulting to bzip2 (current
behavior) if no alternative compression method is specified.
___
freebsd-ports@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-ports
To unsubscribe, send any mail to freebsd-ports-unsubscr...@freebsd.org


Re: packages compressed with xz

2010-11-29 Thread Matthias Andree
Am 28.11.2010 22:12, schrieb Goran Tal:
 Now that the base system supports xz compression, it should be used as
 the default compression for packages.
 
 Files compressed with xz are smaller and decompress faster than those
 compressed with bzip2. This can make an installation much quicker,
 especially when the complete system is installed or upgraded.
 
 Any reasons against it?

xz compressed files can take up CONSIDERABLY more memory to decompress
than files compressed with bzip2 or gzip.  Keep that in mind so that
systems that are low on memory can still decompress xz packages.  If you
don't fit into RAM for decompression, it will be unusable.
___
freebsd-ports@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-ports
To unsubscribe, send any mail to freebsd-ports-unsubscr...@freebsd.org


Re: packages compressed with xz

2010-11-29 Thread jhell
On 11/29/2010 06:24, Matthias Andree wrote:
 Am 28.11.2010 22:12, schrieb Goran Tal:
 Now that the base system supports xz compression, it should be used as
 the default compression for packages.

 Files compressed with xz are smaller and decompress faster than those
 compressed with bzip2. This can make an installation much quicker,
 especially when the complete system is installed or upgraded.

 Any reasons against it?
 
 xz compressed files can take up CONSIDERABLY more memory to decompress
 than files compressed with bzip2 or gzip.  Keep that in mind so that
 systems that are low on memory can still decompress xz packages.  If you
 don't fit into RAM for decompression, it will be unusable.

Adding to this, as the manual says... The decompressing host will need
to have at minimal 5% - 20% of memory 'available' for decompression of
what the compressing host had. Seeing as FreeBSD still runs on systems
with memory as little as 200MB ~20% of 1024MB and quite possible to
run on systems with memory of 64MB ~5% of 1024MB I would not see any
benefit in modifying the default memory limit on a compressing host to
accommodate for these system rather than using gzip(1) or bzip2(1) by
default.

It would be nice to support xz(1) compression for large selective
packages like firefox or openoffice as those will never run on smaller
systems.

-- 

 jhell,v
___
freebsd-ports@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-ports
To unsubscribe, send any mail to freebsd-ports-unsubscr...@freebsd.org


Re: packages compressed with xz

2010-11-29 Thread Ion-Mihai Tetcu
On Mon, 29 Nov 2010 13:31:10 -0500
jhell jh...@dataix.net wrote:

 On 11/29/2010 06:24, Matthias Andree wrote:
  Am 28.11.2010 22:12, schrieb Goran Tal:
  Now that the base system supports xz compression, it should be
  used as the default compression for packages.
 
  Files compressed with xz are smaller and decompress faster than
  those compressed with bzip2. This can make an installation much
  quicker, especially when the complete system is installed or
  upgraded.
 
  Any reasons against it?
  
  xz compressed files can take up CONSIDERABLY more memory to
  decompress than files compressed with bzip2 or gzip.  Keep that in
  mind so that systems that are low on memory can still decompress xz
  packages.  If you don't fit into RAM for decompression, it will be
  unusable.
 
 Adding to this, as the manual says... The decompressing host will need
 to have at minimal 5% - 20% of memory 'available' for decompression
 of what the compressing host had. Seeing as FreeBSD still runs on
 systems with memory as little as 200MB ~20% of 1024MB and quite
 possible to run on systems with memory of 64MB ~5% of 1024MB I
 would not see any benefit in modifying the default memory limit on a
 compressing host to accommodate for these system rather than using
 gzip(1) or bzip2(1) by default.
 
 It would be nice to support xz(1) compression for large selective
 packages like firefox or openoffice as those will never run on smaller
 systems.

Trouble is it ain't no way (CPU, space, banhdwidth on our side and
space,bandwidth on our mirrors side) we could build a double set of
packages.

-- 
IOnut - Un^d^dregistered ;) FreeBSD user
  Intellectual Property is   nowhere near as valuable   as Intellect
FreeBSD committer - ite...@freebsd.org, PGP Key ID 057E9F8B493A297B


signature.asc
Description: PGP signature


Re: packages compressed with xz

2010-11-29 Thread Matthias Andree
Am 29.11.2010 19:31, schrieb jhell:

 Adding to this, as the manual says... The decompressing host will need
 to have at minimal 5% - 20% of memory 'available' for decompression of
 what the compressing host had. Seeing as FreeBSD still runs on systems
 with memory as little as 200MB ~20% of 1024MB and quite possible to
 run on systems with memory of 64MB ~5% of 1024MB I would not see any
 benefit in modifying the default memory limit on a compressing host to
 accommodate for these system rather than using gzip(1) or bzip2(1) by
 default.

You can specify limits during compression, so the question is should we do that
so that hosts with N MB of RAM can decompress packages?  Do we retain the
compression ratio over bzip2 if we limit compression memory to 512 MB so that
decompression would be possible with, say, 128 MB?

 It would be nice to support xz(1) compression for large selective
 packages like firefox or openoffice as those will never run on smaller
 systems.

Yes, would be nice.  I doubt it will happen soon.

-- 
Matthias Andree
___
freebsd-ports@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-ports
To unsubscribe, send any mail to freebsd-ports-unsubscr...@freebsd.org


Re: packages compressed with xz

2010-11-29 Thread Mark Linimon
On Tue, Nov 30, 2010 at 12:40:33AM +0100, Matthias Andree wrote:
 Yes, would be nice.  I doubt it will happen soon.

It's actually being looked at.

As part of the extensive rework of the pointyhat scripts I did this
summer, I attempted to factor out all the magic constants, including
the definition of PKGSUFFIX.

There are some other bugs that need to be fixed (having to deal with
INDEX and duds) before we can get to this, however.

mcl
___
freebsd-ports@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-ports
To unsubscribe, send any mail to freebsd-ports-unsubscr...@freebsd.org


Re: packages compressed with xz

2010-11-29 Thread jhell
On 11/29/2010 18:40, Matthias Andree wrote:
 Am 29.11.2010 19:31, schrieb jhell:
 
 Adding to this, as the manual says... The decompressing host will need
 to have at minimal 5% - 20% of memory 'available' for decompression of
 what the compressing host had. Seeing as FreeBSD still runs on systems
 with memory as little as 200MB ~20% of 1024MB and quite possible to
 run on systems with memory of 64MB ~5% of 1024MB I would not see any
 benefit in modifying the default memory limit on a compressing host to
 accommodate for these system rather than using gzip(1) or bzip2(1) by
 default.
 
 You can specify limits during compression, so the question is should we do 
 that
 so that hosts with N MB of RAM can decompress packages?  Do we retain the
 compression ratio over bzip2 if we limit compression memory to 512 MB so that
 decompression would be possible with, say, 128 MB?
 

Hosts that have [N]MB of free or available memory. Most systems in
this case will not have a whole lot of RAM available in any case as they
are likely to be utilizing it to its upper most potential. Doing such
limiting on the compressors part I would think, be more of a waste of
resources as such can be achieved by default just sticking with bzip2(1).

Besides, limiting memory to 512MB to what ? shave an ~ small percentage
off the top of a resulting package.

 It would be nice to support xz(1) compression for large selective
 packages like firefox or openoffice as those will never run on smaller
 systems.
 
 Yes, would be nice.  I doubt it will happen soon.
 

Agreed. Soon can be quantified by actual need and of which there is not
much need except for larger packages but adding this would just add
unneeded complication to the system that is already in place.

~ JMO

-- 

 jhell,v
___
freebsd-ports@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-ports
To unsubscribe, send any mail to freebsd-ports-unsubscr...@freebsd.org


Re: packages compressed with xz

2010-11-29 Thread Julien Laffaye
 On 11/29/2010 23:40, Matthias Andree wrote:

You can specify limits during compression, so the question is should we do that
so that hosts with N MB of RAM can decompress packages?  Do we retain the
compression ratio over bzip2 if we limit compression memory to 512 MB so that
decompression would be possible with, say, 128 MB?

According to xz(1), in its default mode (-6), xz uses ~100MiB for
compression and ~10MiB for decompression.
That seems to be acceptable.
___
freebsd-ports@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-ports
To unsubscribe, send any mail to freebsd-ports-unsubscr...@freebsd.org


packages compressed with xz

2010-11-28 Thread Goran Tal
Now that the base system supports xz compression, it should be used as
the default compression for packages.

Files compressed with xz are smaller and decompress faster than those
compressed with bzip2. This can make an installation much quicker,
especially when the complete system is installed or upgraded.

Any reasons against it?
___
freebsd-ports@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-ports
To unsubscribe, send any mail to freebsd-ports-unsubscr...@freebsd.org


Re: packages compressed with xz

2010-11-28 Thread Dominic Fandrey
Hello,

On 28/11/2010 22:12, Goran Tal wrote:
 Now that the base system supports xz compression, it should be used as
 the default compression for packages.
 
 Files compressed with xz are smaller and decompress faster than those
 compressed with bzip2. This can make an installation much quicker,
 especially when the complete system is installed or upgraded.
 
 Any reasons against it?

I think it's already in CURRENT.

On my notebook xz compression and decompression are much slower
than any of the other algorithms supported by tar. It has the
best compression ratio, though.
___
freebsd-ports@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-ports
To unsubscribe, send any mail to freebsd-ports-unsubscr...@freebsd.org


Re: packages compressed with xz

2010-11-28 Thread Ion-Mihai Tetcu
On Sun, 28 Nov 2010 16:12:48 -0500
Goran Tal goran@gmail.com wrote:

 Now that the base system supports xz compression, it should be used as
 the default compression for packages.
 
 Files compressed with xz are smaller and decompress faster than those
 compressed with bzip2. This can make an installation much quicker,
 especially when the complete system is installed or upgraded.
 
 Any reasons against it?

Sigh. We enabled that in HEAD, and after some testing and adapting
{base_OS, ports, tools, etc.} to support that, it will probably become
the default.

One thing against it is the major increase in package time (up to 6x,
but by no means this is a show-stopper).
Resulting packages would be, in medium about 30% smaller (in medium,
for some kinds of data the size is the same as for gzip).

So no, there's no reason against it, just some time is need to adapt
everything.

-- 
IOnut - Un^d^dregistered ;) FreeBSD user
  Intellectual Property is   nowhere near as valuable   as Intellect
FreeBSD committer - ite...@freebsd.org, PGP Key ID 057E9F8B493A297B


signature.asc
Description: PGP signature