Jan Setje-Eilers wrote:
> Alexander Eremin wrote:
>> On ??, 2009-10-22 at 10:56 +0200, Jan Damborsky wrote:
>>> Hi Alex,
>>>
>>> since Sparc uses dcfs(7FS) "Compression file system" module
>>> to decompress files on-the-fly when data are being read, my
>>> understanding
>>> is that all pieces necessary to bring up the system to the point when
>>> dcfs(7FS) can do its job must remain uncompressed
>>>
>>> I am not sure what the critical pieces are - CC-in Jan S. who
>>> might correct me or provide more details on this topic.
>>>
>>> That said, I am thinking how much space we actually save in this case
>>> (Sparc OpenSolaris installer boot archive) - '/usr' is separated
>>> in compressed archive solaris.zlib and kernel along with drivers
>>> is uncompressed.
>>>
>>> If it turns out it is not significant, I think we might also consider
>>> not to use dcfs at all, since there are also disadvantages - we are
>>> seeing weird bugs filed from time to time against Sparc - they
>>> are caused by the limitations of dcfs - compressed file can't be
>>> updated. This is why 'fiocompress' attribute in Distro Constructor
>>> Sparc manifest exists - files marked with this attribute are left
>>> uncompressed.
>>>
>>> Thank you,
>>> Jan
>>
>> I've prepared some Sparc iso's using root_archive for creating miniroot,
>> they work good, but I?m agree about updating question. For DC it seems
>> that only small part of bootroot is fiocompressed - /sbin,/usr and /libs
>> and probably using dcfs for miniroot compression is not optimal it this
>> case, other method is required.
>
> I agree that the win may not be worth the trade-offs when /usr is
> compressed in another way.
Thanks, Jan !
Alex,
I tried a quick test and created uncompressed Sparc boot archive
from the compressed one:
# mkdir /tmp/br /tmp/br1
# mount -o nologging
/export/home/ai/images/1003s-ai-12200-125/boot/boot_archive /tmp/br/
# /usr/sbin/mkfile 170m /boot_archive_nofiocompress
# /usr/sbin/lofiadm -a /boot_archive_nofiocompress
/dev/lofi/2
# /usr/sbin/newfs /dev/lofi/2
# /usr/sbin/mount -o nologging /dev/lofi/2 /tmp/br1/
# rsync -av /tmp/br/ /tmp/br1/
...
# umount /tmp/br1
# lofiadm -d /dev/lofi/2
# mount -o nologging /boot_archive_nofiocompress /tmp/br1
Original compressed archive:
# df -h /tmp/br/
Filesystem Size Used Avail Use% Mounted on
/export/home/ai/images/1003s-ai-12200-125/boot/boot_archive
152M 121M 32M 80% /tmp/br
Uncompressed archive:
# df -h /tmp/br1/
Filesystem Size Used Avail Use% Mounted on
/boot_archive_nofiocompress
160M 151M 0 100% /tmp/br1
So the difference is ~30MB with respect to 'Used' space.
Comparing total size, the difference could be even smaller,
since there is ~32MB of free space in current compressed.
As Karen pointed out, 10% of free space is the value we
should calculate with which would be ~15MB.
Based on this, I am thinking if we could completely omit
Sparc archive compression step and remove it from DC.
pros:
* faster generation of Sparc AI image
* simplified build procedure
* simplified manifest (no need for list of 'fiocompress=false' files)
* no need to worry about issues coming from dsfs(7FS) limitations
cons:
* bigger Sparc boot archive => bigger AI image (~15MB estimated difference)
Thank you,
Jan
