[uClibc]Re: Putting uclibc in a linux distribution?

Rob Landley landley at trommello.org
Thu Sep 12 16:07:16 UTC 2002


On Thursday 12 September 2002 02:38 am, Christian MICHON wrote:

>  * zisofs is cool, but it creates a node for each file. The advantage of
> cloop for similar performances but *much* smaller size in a single cloop
> file.

Huh?  (How do you avoid creating an inode for each file...?)  I've never 
heard of cloop.  Then again google could find jimmy hoffa if they could just 
figure out a way to hook it up to yahoo maps...

http://www.knopper.net/download/knoppix/cloop.README

Interesting.  I think what you mean is that each file is compressed 
individually (like a zip file instead of a gzipped tarball).  This doesn't 
provide as good overall compression, but in terms of random access to a bunch 
of small files, it's a good thing.  (It's hard to seek around in a compressed 
file, you have to backup to the last dictionary reset and decompress forward 
from there.)

I don't know the implementation details (I'll have to take a look at it), but 
I'm curious how he avoided the "loopback mounting a tar file just doesn't 
work" problem.  (He's got to be keeping an index of dictionary reset 
positions...)

Thanks for the pointer, but right now this isn't part of the standard kernel 
and zisofs is (as is cramfs, which doesn't work.  I think it tries to open 
every file to detect hard links, and runs out of filehandles before it's 
halfway through the root partition).  (On a side note, I have looked into the 
crypto loopback mount stuff for the /home and /var partitions, but there's 
enough politics there to  keep it pretty far down on my to-do list.  I am 
curious if the two patches interoperate, though...)

> Now that most of the problems above are fixed, I'll give it a shot too. ;)

The more the merrier... :)

> Christian

Rob



More information about the uClibc mailing list