[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: md (software) RAID missing in the installer for sparc



On Tue, 17 Apr 2007, Josip Rodin wrote:

4) If the partition table on the disk was first written by an old
   version of the Solaris format command the maximum useable size of
   any partition will be either 1Gb or 2Gb depending on version.
   Linux will probably not notice and allow you to create a larger
   partition which will confuse the OBP even if the OBP does not have
   the limitation in 2) above.
   The workaround for this is to destroy cylinder zero by filling it
   with zeros using dd and then use Linux fdisk to create a new BSD
   table.  [ Affects disks initially formatted with old utilities ]

I had Solaris on it previously, but I only had Linux parted and fdisk
trample over it several times, I never filled it with zeros. Is that really
necessary? :)

Yes, The table is only created when the disk is first labelled.
The table does not change, only the entries in the table.
The table contains version information which must be preserved because
the data format changed between versions to allow larger pointers.
The same problem also shows up on IDE disks, if older utilities were
used to label the disk it will always have a maximum size of 8Gb or 37Gb
until you zap the table and make a new one.

There is also a problem with SILO which I do not fully understand, but
it can also have problems looking more than 2047 Mb into a partition.
This can sneak up on you if you edit any of SILO's files, the new file
can be past the limit and make the system unbootable.

It is always best practice to have a small /boot partition at the
beginning of the disk.

BTW - If you are using RAID for your data you should also mirror swap.

Do not use anything other than a simple mirror for swap and do not have
more than one swap area on the same spindle. Both Solaris and Linux will
try to optimise performance by striping swap data across as many swaps
as you have, so having more than one swap per spindle will guarantee
serious head thrashing.

Maybe someone more knowledgeable will explain the SILO problems.




Reply to: