24 TB UFS2 reality check ?

[Julian Elischer]

Juri Mianovich wrote:
> 
> 
> --- On Tue, 7/8/08, Julian Elischer <julian at elischer.org> wrote:
> 
> 
>> You had better have a lot of memory available ot your
>> processes to be 
>> able to fsck this baby.. (it'd better be an amd64)..
>> I don't remember the exact numbers but for 16k
>> blocksize,
>> it was something like 200MB ram for each 100GB of
>> filesystem when 
>> populated with 60KB files..
>> (don't trust those numbers, do some testing (and let us
>> know :-) )
> 
> 
> Thank you very much.
> 
> I currently have a similar system with:
> 
> /dev/da1       8.0T    1.3T    6.1T    16%    /users
> 
> which was created with 'newfs -i 32768 -U /dev/da1' ... and I can
> successfully fsck it with my:
> 
> kern.maxdsiz="2572000000"
> 
> setting.
> 
> So perhaps a filesystem 3x that size should be '-i 131072' to
> maintain the same ability to fsck ?
> 
> None of these systems are 64-bit - they are all running i386 w/4 GB
> of ram.
> 
> If I stuck with '-i 65536' (instead of going all the way to 131072)
> and things got sticky, I could always temporarily reboot with a
> maxdsiz closer to 3 GB, right ?

You will have to do tests to see how big the virtual size of
the fsck process gets per TB of disk.
you should be able to see it with top.
Notice that your 8TB filesystem is only 16% full. The memory will
increase to some exgtent when you have more files..
try filling up your 8TB system with files and doing it again.

> 
>