"Jamey Hicks" <jamey@crl.dec.com> writes:
> I find
> that cramfs uses half the space as ext2 for the same files.
On the ARM? Consider yourself lucky. I'm getting a compression factor of
2.7 on MIPS, which sounds great until you realize that at some point the
code has to be uncompressed. SVR4 MIPS ABI is just amazingly bloaty.
> We use a variety of strategies on Itsy, Skiff, and the iPaq. On all of
> these, we usually store the zImage in a flash partition and the firmware
> extracts and runs it. Most of the Skiffs are running with zImage copied
> from a flash partition and NFS-mounted root. On the iPaq, we either
> cramfs/flash or ext2/ramdisk for the root filesystem with other ext2 or
> cramfs filesystems mounted on the flash block driver.
If there's a lot of flash and its access speed isn't too awful, it would be
nice to be able to execute-in-place code from flash. Obviously you can't do
this for cramfs; I haven't looked at JFFS enough to see if XIP is possible
on any layered filesystem. XIP isn't possible for the romdisk.c patch
floating around.
Under cramfs, uncompressed pages that aren't in use are candidates for being
dropped. So if you do XIP you won't see free RAM go up by the amount of
storage moved to flash, because some of the savings are already there.
I'm starting to fret about code density/coverage. We only have to keep a
page from cramfs swapped in when it has executed code on it; it would be
nice to make sure that page has other functions that are actually in use.
There's some info on an open source replacement for cord, called GNU Rope at
http://lwn.net/1998/1029/als/rope.html but there's no sign of the code on
the net. I'm going to try mailing the author, but I suspect they're pretty
busy right now.
> With a cramfs root,
> you'll still want to use ext2/ramdisk to give the system some writable tmp
> and var space.
Have you tried out ramfs yet? I haven't, but it sounds like it will perform
better than a disk fs on a block ramdisk device.
Jay
Received on Thu Jun 29 06:37:26 2000
This archive was generated by hypermail 2.1.8 : Tue May 04 2004 - 09:42:15 EDT