On 27 April 2017 at 22:42, Björn 'besser82' Esser <besser82@xxxxxxxxxxxxxxxxx> wrote:
> Fedora currently has about 19k packages in each maintained release / branch,
> thats a total of about 76k packages. If such a little change saves about
> 100KB of storage in average for every single package, it's a total saving of
> about 7.25 GB. This saves about 7.25 GB for syncing of each mirror; for 242
> active mirrors that's a total of 1.71 TB of transmitted data…
bash provides in own doc subpackages tripled the same documentation (info and two copies of the pdf version the same doc in files under different names).
Why no one cares abou this?
Try to think that if packages will exist only as the objects in repository it will be not possible to save 1GB but much more by sharing on the repos server side common parts between packages for different architectures.
Try to have look for example on mc package metadata in Solaris IPS repository http://pkg.oracle.com/solaris/release/manifest/0/file%2Fmc@4.8.17%2C5.12-5.12.0.0.0.115.0%3A20170111T171112Z
You can find in those metadata lines:
file 7080b660d77d5d260f95e269626f2900b36cba0e chash=c2ef7983c385175c32f6ac6990175289529aa1ea elfarch=sparc elfbits=64 elfhash=0fd93599aef1c81668aad3063a2ddc2d8fa60fd6 group=bin mode=0555 owner=root path=usr/bin/mc pkg.csize=733172 pkg.size=2059216 variant.arch=sparc
and
file be26b559c366018c655c16d3dabf46ab61bcf962 chash=5e2285f84f85e71e63f763ea151ecabed9e342b3 elfarch=i386 elfbits=64 elfhash=c171f543984c3da03cf479dbc10d1ec3b0bf9aac group=bin mode=0555 owner=root path=usr/bin/mc pkg.csize=622533 pkg.size=1916544 variant.arch=i386
One package holds all architectures binaries sharing all other files.
Do you see now how much MORE on all packages servers is possible to save disk space using this approach?
In IPS repository is not a problem keeping longer history of different versions or the same package across many revisions/versions as most of the files on repo server side will be shared between those multiple versions. Do you see this now?
In Fedora for me most frequent problem is that as long as I'musing rawhide sometimes I'm not doing regularly every day updates and when I'm finding that some process is crashing I cannot download correct debuginfo packages because it already has been deleted.
In publically available IPS repos like http://pkg.oracle.com/solaris/release/en/index.shtml you ca find only few major releases. However with support you can have access to all past every ~month SRUs packages revisions from all last few years and it does not need tenths of terabytes of storage.
In Solaris each binary is served with way smaller deguginfo integrated into regular binaries. Size of this additional data is so small that Solaris developers decided to provide debuginfo in regular distro packages.
In Fedora debuginfo are sometimes 10x bigger than rest of the regular package.
If you really care about size here are biggest deposits of things which Linux can learn from his more mature cousins.
Solaris IPS was reply on RPM and growing demands of solving other problems which Fedora barely stats to fight (try to have look on IPS mediators and you will cee that they are solves ALL problems which Modules guys just started thinking how to solve on top of unchanged RPM).
After IPS introduction about 9 years ago nothing happen on RPM area .. full stagnation which now causes that some desperate package developers are thinking about create even more subpackages because RPM cannot provide some useful solutions.
Guys if you are thinking that splitting langpack subpackages may solve something you are 100% right .. you are only thinking.
Why no one cares abou this?
Try to think that if packages will exist only as the objects in repository it will be not possible to save 1GB but much more by sharing on the repos server side common parts between packages for different architectures.
Try to have look for example on mc package metadata in Solaris IPS repository http://pkg.oracle.com/solaris/release/manifest/0/file%2Fmc@4.8.17%2C5.12-5.12.0.0.0.115.0%3A20170111T171112Z
You can find in those metadata lines:
file 7080b660d77d5d260f95e269626f2900b36cba0e chash=c2ef7983c385175c32f6ac6990175289529aa1ea elfarch=sparc elfbits=64 elfhash=0fd93599aef1c81668aad3063a2ddc2d8fa60fd6 group=bin mode=0555 owner=root path=usr/bin/mc pkg.csize=733172 pkg.size=2059216 variant.arch=sparc
and
file be26b559c366018c655c16d3dabf46ab61bcf962 chash=5e2285f84f85e71e63f763ea151ecabed9e342b3 elfarch=i386 elfbits=64 elfhash=c171f543984c3da03cf479dbc10d1ec3b0bf9aac group=bin mode=0555 owner=root path=usr/bin/mc pkg.csize=622533 pkg.size=1916544 variant.arch=i386
One package holds all architectures binaries sharing all other files.
Do you see now how much MORE on all packages servers is possible to save disk space using this approach?
In IPS repository is not a problem keeping longer history of different versions or the same package across many revisions/versions as most of the files on repo server side will be shared between those multiple versions. Do you see this now?
In Fedora for me most frequent problem is that as long as I'musing rawhide sometimes I'm not doing regularly every day updates and when I'm finding that some process is crashing I cannot download correct debuginfo packages because it already has been deleted.
In publically available IPS repos like http://pkg.oracle.com/solaris/release/en/index.shtml you ca find only few major releases. However with support you can have access to all past every ~month SRUs packages revisions from all last few years and it does not need tenths of terabytes of storage.
In Solaris each binary is served with way smaller deguginfo integrated into regular binaries. Size of this additional data is so small that Solaris developers decided to provide debuginfo in regular distro packages.
In Fedora debuginfo are sometimes 10x bigger than rest of the regular package.
If you really care about size here are biggest deposits of things which Linux can learn from his more mature cousins.
Solaris IPS was reply on RPM and growing demands of solving other problems which Fedora barely stats to fight (try to have look on IPS mediators and you will cee that they are solves ALL problems which Modules guys just started thinking how to solve on top of unchanged RPM).
After IPS introduction about 9 years ago nothing happen on RPM area .. full stagnation which now causes that some desperate package developers are thinking about create even more subpackages because RPM cannot provide some useful solutions.
Guys if you are thinking that splitting langpack subpackages may solve something you are 100% right .. you are only thinking.
--
Tomasz Kłoczko | Tel: 0774 1209067 | LinkedIn: http://lnkd.in/FXPWxH
_______________________________________________ devel mailing list -- devel@xxxxxxxxxxxxxxxxxxxxxxx To unsubscribe send an email to devel-leave@xxxxxxxxxxxxxxxxxxxxxxx
