disk partitioning "future safety" option

[dhylton]

background:

not all disk sizes are reported equally across manufacturers, and sometimes not even within a given model. over the years, i've been bitten multiple times when replacing failed disks that were members of an array; sometimes, replacing a failed 200GB disk inside a raid1 array with another 200GB disk will fail because the replacement has a few less sectors than the original. this usually results in replacement with a larger disk (eg 250GB), after the required colorful language and a second trip to the parts store. oh, and an unused but unreturnable 200GB disk.

while installing bsd (or most other systems) in the normal fashion, and for the addition of any new partitions/volumes, i typically create and use a slice which is slightly smaller than the whole device. that protects me from the replacement-disk-is-slightly-smaller problem mentioned above. and it works.

does anyone else do this, or am i alone?


request/suggestion:

it'd be great for the installer to provide a checkbox that would allow me to choose for the installer to perform this little safety dance for me automatically. otherwise i can do the partitioning myself, but it's obviously more difficult and error-prone for the user to do so manually - and then we lose out on the intelligent defaults for the automatically-created filesystems (eg the huge list of zfs filesystems with varying flags).

[Tigersharke]

I wonder if part of the problem relates to base-10 size versus base-2 size, though whether it does or not is another issue. This checkbox to reserve space on the HDD seems like a reasonable request, but how much should it reserve?

• a set percent of the entire drive?
• <drive size in base-2> minus <drive size in base-10>?
• A set amount in bytes, blocks, or MB?
• According to a sector boundary?

[dhylton]

great questions. i've never researched the differences to see whether or not there was a discernable pattern. the differences i've seen have been fairly miniscule - just enough to cause the failure (and to trigger colorful language). perhaps the checkbox could enable a text box that would allow the user to define the number of sectors/cylinders/megabytes to be reserved.

[Tigersharke]

Hmm.. Ok. I noticed this "do not use your whole HDD" post, which also says how some larger drives can vary in actual size.

So, rather than say "reserve X" via checkbox or slider or whatever method, I think the real idea is more like "define X as whole disk size" where default is 100% of actual size, and can be reduced to arbitrary amounts, leaving unused space. In the case of ZFS as mentioned in the article above, it allows for precise and equally sized HDDs, apart from what the manufacturer(s) provide(s).

Also, if it is possible to use *whole* disk, then the above as a variation of this should also be possible, but the key being that the unused space is purely unused and always unused (ie. not recovered at a later date for NTFS or some other purpose). In the context of ZFS with its constraints on arrays (I am very new to ZFS so I mostly assume at this point) it makes sense.

[dhylton]

thanks for pointing to that post! that person has seen exactly what i'm trying to get around. and the issue has nothing to do with zfs, really; the same problem happens with hardware raid controllers as well. one simply cannot mirror a given disk to a smaller device (same applies to other raid types).

so ... there are obviously hardware inconsistencies over which we have no control, but we do have the ability to create a software workaround. i do this manually, for several operating systems, today. i'm hoping that kris and company might see this as a good idea and build it into the installer. and perhaps freebsd will do something with it as well.