Jan Damborsky wrote:
> Hi Alex,
>
> I have given those changes a try and while DC build succeeded
> for Sparc and x86 AI images, number of available inodes
> were insufficient for nbpi=16384 in x86 LiveCD case for both
> 32 as well as 64 bit archives.
>
> So I have taken a closer look at boot archives based on build 128 -
> observations are summarized in tables below.
>
> From this it seems that 16384 is sufficient for AI images, 8192
> is needed in LiveCD case.
>
> I am thinking about following solutions:
>
> [1] Specify nbpi in DC manifests.
> Looking at DC manifest, there is already section <boot_archive>
> which is used to set boot archive parameters - e.g.
> compression type (<compression>) or padding (<size_pad_mb>).
> We could add support for nbpi.
>
> [2] Work out algorithm which would automatically calculate nbpi
> during process of creating boot archive.
> For instance, nbpi might be function of ramdisk size and
> number of inodes:
> nbpi = f(size of backend file, number of required inodes)
>
> [3] ?
>
>
> I think mentioned approaches have their pros and cons, e.g.
>
> [1]
> + user can set it to the desired value
> - if size or content of boot archive changes in such a way that
>  build of boot archive fails, manual intervention is required -
>  nbpi has to be recalculated
>
> [2]
> + no need to calculate the value manually
> - less control - user can't set it to the desired value
>
> Please let me know what you think.
Since the nbpi is a way to tune the size of the boot archive,
I think that we should provide a way for users to
specify different values so they can try out what works best for
the problem they are solving.

I think we can provide both option 1 and 2 you mentioned
above.  By default, have the value be calculated automatically, and display
what value is used in the log file.  If users want to try out a 
different value,
they can specify the number to use in the manifest, and that number will
be used, and no automatic calculation is done for them.

Thanks,

--Karen
>
> Thank you,
> Jan
>
>
> Tables:
>
> [1] size of ramdisk backend file [kB] - uncompressed in case of x86
>
> For instance for AI x86 64 bit archive the size was obtained using
> following steps:
>
> # gzcat  /tmp/iso_ai/platform/i86pc/amd64/boot_archive > 
> /tmp/boot_archive_ai_64
> # du -sk /tmp/boot_archive_ai_64
> 159400    /tmp/boot_archive_ai_64
> #
>
> -----------------------------------------
>  cd_32    cd_64  ai_32  ai_64  ai_sparc
> -----------------------------------------
> 144452   177132 114256  159400   170968
> -----------------------------------------
>
>
>
> [2] # of used inodes
>
> fsck(1M) was used to obtain this information, e.g.
> # fsck -v /tmp/boot_archive_ai_64
> ...
>
> Filesystem summary:
> Inodes in use: 6977
> Blocks in use: 128994
> Total free fragments: 28385
> Free fragments not in blocks: 225
> Total free blocks: 3520
> Fragment/block fragmentation: 0.1%
>
>
> ----------------------------------------
> cd_32    cd_64  ai_32  ai_64  ai_sparc
> ----------------------------------------
> 14892    15095   6775   6977      7310
> ----------------------------------------
>
> [3] # of available inodes
>
> Obtained experimentally - filesystem was filled up
> with 1byte files.
>
> -----------------------------------------------
> nbpi   cd_32    cd_64  ai_32  ai_64  ai_sparc
> -----------------------------------------------
> 8192   17854    21286      -      -         -
> 16384    9918    11838   7678  10878     11518
> -----------------------------------------------
>
> Legend:
> cd_32     x86 LiveCD 32bit boot archive
> cd_64     x86 LiveCD 64bit boot archive
> ai_32     x86 AI 32bit boot archive
> ai_64     x86 AI 64bit boot archive
> ai_sparc  Sparc AI boot archive
>
>
>>> Alex,
>>> you are quite welcome.
>>>
>>> Could you please send me your changes by means of 'hg
>>> bundle' ?
>>>
>>> Thank you,
>>> Jan
>
>>> Done.
>>>
>>> -- 
>>> Alexander
>
> _______________________________________________
> caiman-discuss mailing list
> caiman-discuss at opensolaris.org
> http://mail.opensolaris.org/mailman/listinfo/caiman-discuss


Reply via email to