Centralized repository

Centralized repository

[bob]

I know that this has been mentioned before.  So, I will only bring it  
up once (from me at least).

I have a project that is currently over 2gigs without .git and  
growing.  It is a personal website that I use to provide/remind me of  
things that I used to remember when I was younger.  When you put a  
GIT repostitory in it, you get almost 4gigs of space taken (and this  
amount is growing all the time).  Then I keep it centralized, because  
I own several machines (OK, way too many, over 20 currently,  
dedicated to Folding@Home).  So, it takes up a lot of disk space over  
all those machines.

Because it is so big, cloning no longer works.  I can't remember the  
exact errors, but I gave up and now tar up  a semi-current version  
and untar it on a new/re-installed machine.  Then I do a "pull" to  
update it and that works when the clone doesn't.  I should mention  
that I have only 2 windows machines, but run GIT on all the rest  
which are Linux and MacOSX machines.

Anyway, for this one repository, I would really like GIT to support  
the CVS/SVN central repository system.  The rest of the time, I am  
quite happy with GIT's distributed system even though all of my  
repositories are really just for me.

Last, thanks for GIT.  I really enjoy it and use it daily and have  
never had it fail on me (no matter what I did!)!

Bob White

Re: Centralized repository

[Jeff King]

On Sat, Feb 24, 2007 at 09:14:27PM -0500, bob wrote:

> Anyway, for this one repository, I would really like GIT to support  
> the CVS/SVN central repository system.  The rest of the time, I am  
> quite happy with GIT's distributed system even though all of my  
> repositories are really just for me.

For a while I was running git with an 80-gig repository of media files.
I found it useful to network mount the central git repo (I used SMB),
and clone to the local machine using alternates (i.e., "git clone -s
/net/mounted/git /local/git").  It should work as a normal git repo,
except that all object accesses will transparently go over the network,
which obviously slows things down a bit (but no slower than CVS!).

You should be able to pack at your size of repo (I obviously couldn't,
but I was working with a few very large files, so the overhead wasn't
too bad).

HTH,
-Peff

Re: Centralized repository

[Andy Parkins]

On Sunday 2007, February 25, bob wrote:

> Anyway, for this one repository, I would really like GIT to support
> the CVS/SVN central repository system.  The rest of the time, I am

In the case of subversion: you really /wouldn't/ want that model.  
Subversion keeps a local pristine copy in the .svn directory of every 
file.  This makes local diffs and stats fast and not require network 
access, but also doubles the storage requirements of your working 
directory.

I've found that a packed git repository is often smaller than a full 
checkout of the project (it's almost scary actually - your entire 
project's history in less space than the project).


Andy
-- 
Dr Andrew Parkins, M Eng (Hons), AMIEE
andyparkins@gmail.com

Re: Centralized repository

[Robin Rosenberg]

söndag 25 februari 2007 10:51 skrev Andy Parkins:
> On Sunday 2007, February 25, bob wrote:
> 
> > Anyway, for this one repository, I would really like GIT to support
> > the CVS/SVN central repository system.  The rest of the time, I am
> 
> In the case of subversion: you really /wouldn't/ want that model.  
> Subversion keeps a local pristine copy in the .svn directory of every 
> file.  This makes local diffs and stats fast and not require network 
> access, but also doubles the storage requirements of your working 
> directory.
> 
> I've found that a packed git repository is often smaller than a full 
> checkout of the project (it's almost scary actually - your entire 
> project's history in less space than the project).

That doesn't apply to a repo full of (already compressed) binaries media 
files. You cannot pack them unless the binaries are so small  that the wasted 
space at the end of the last disk block is a significant amount of the total 
used. You cannot diff them either, other than seeing that they are different. 

Besides the 4 GB mentioned isn't much nowadays. If it is annoying to have 
the .git directory in the same filesystem tree you can place the .git 
directory somewhere else and drop a  symbolic link in it's place. 

Btw, is it possible to make a full repo shallow again, by pruning  
the "uninteresting" part of the history? That would make it possible to trim 
down the size of the local archive. 

-- robin

Re: Centralized repository

[Martin Waitz]

[-- Attachment #1: Type: text/plain, Size: 398 bytes --]

hoi :)

On Sat, Feb 24, 2007 at 09:14:27PM -0500, bob wrote:
> Because it is so big, cloning no longer works.  I can't remember the  
> exact errors, but I gave up and now tar up  a semi-current version  
> and untar it on a new/re-installed machine.  Then I do a "pull" to  
> update it and that works when the clone doesn't.

this is fixed in the master branch now.

-- 
Martin Waitz

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

Re: Centralized repository

[bob]

Thank you, Andy Parkins, Jeff King and Robin Rosenberg, for your  
replies.  You have given me some ideas to look into.  I had not  
thought about linking the .git directory to somewhere else such as  
SMB, NFS, AFP or an external harddrive.  I copied a simpler .git to a  
FAT32 drive, made a soft link to it and everything seemed to work  
fine.  I will continue to play.  Thanks for the ideas. They were just  
what I was looking for.