On Mon, Oct 24, 2005 at 01:53:16PM +0100, Julian Coleman wrote:
> > 2.  The reason for choosing a particular dumpdev partition size is not
> >     stated in the RAIDframe documentation and I believe it should be.
> >     It is stated in swapctl(8); however, I believe that there are some
> >     users who will not know how to properly select a size for this
> >     partition and may select one which is too small.  I believe the
> >     RAIDframe documentation should be modified to state that the
> >     dumpdev partition size should be equal to or larger than than the
> >     amount of physical memory the computer has installed.
> 
> This is the case with or without raid.  However, perhaps the RAIDframe
> section could refer to other sections/manual pages?


Julian,

You're right that it is applicable with or without RAID.  I believe
the RAIDframe documentation should at least indicate how to select the
correct size for the dumpdev partition by pointing to swapctl(8).


> >     Furthermore, I believe it would be wise to caution users that they
> >     may wish to consider the maximum amount of physical memory their
> >     computer can accommodate and consider the likelihood of an upgrade
> >     that would cause their dumpdev partition to be undersized.  Do I
> >     correctly understand dumpdev size selection?  Should these
> >     concerns be included in the RAIDframe documentation?
> 
> Again, I think this is more general.  However, the one item specific to
> RAIDframe is having separate swap and dump partitions.
> 
> Also, you can save disk space by overlapping the raid swap device and
> the dump device on the underlying disk (I can't remember who pointed
> this out previously).  For example, I have (on a U5):
> 
>   disklabel sd0:
> #        size    offset     fstype [fsize bsize cpg/sgs]
>  a:  35368272         0       RAID                     # (Cyl.      0 -   7505)
>  d:   1050776   4198392       swap                     # (Cyl.    891 -   1113)
> 
>   disklabel raid0:
> #        size    offset     fstype [fsize bsize cpg/sgs]
>  a:   4195328         0     4.2BSD   2048 16384 21872  # (Cyl.      0 -   4096)
>  b:   2098176   4195328       swap                     # (Cyl.   4097 -   6145)
> 
>   fstab:
> /dev/raid0b     none            swap    sw      0       0
> /dev/sd0d       none            dump    dp      0       0
> 
> The offset of raid0c from the beginning of sd0c is 64 sectors, so as long
> as the dump partition starts after (raid0b offset + 63) and ends before
> (raid0b offset + raid0b size + 64), you'll be fine.  You can test that
> you've got it right by using dd to write to the start and end sectors of
> sd0d and to read from the corresponding sectors in raid0b (whilst single
> user).
> 
> J


This is, in fact, the only way documented by the RAIDframe
documentation!  It's nice to know such an efficient use of disk space
is possible, but for the less saavy it might be nice to explain how to
configure RAID-1 without a dumpdev partition and how a user could have
swap and dumpdev partitions which do not overlap.  Do you agree?

Could you summarize for me why someone might choose to have a dumpdev
partition?

Thanks for your feedback,
Ryan