Bug#652275: Guided partitioning should not offer separate /usr, /var, and /tmp partitions; leave that to manual partitioning

[Goswin von Brederlow <goswin-v-b@web.de>]

Russell Coker <russell@coker.com.au> writes:

> On Sun, 18 Dec 2011, Josselin Mouette <joss@debian.org> wrote:
>> > Doing this has many advantage. Like, if your laptop has to unexpectedly
>> > reboot (like when you inadvertently removed power cord when batteries
>> > were not plugged, which happens often in real life), having separated
>> > partitions usually makes the fsck faster.
>> 
>> This is complete bullshit. With a journaled filesystem, the boot time
>> will greatly increase with the number of filesystems to check. If no
>> files were modified in /usr, they wonâ??t be mentioned in the journal, and
>> thatâ??s all. But having one journal to parse for all the system is
>> definitely a measurable improvement.
>
> If we want to improve fsck time then the best thing to do would be to consider 
> a different default value for the -i option of mke2fs.
>
> The current default is to have one Inode per 16K of disk space.  Of the 
> Maildir format mail servers that I run the one with the smallest disk space 
> used per Inode has 307G and 4773821 Inodes in use for an average of 67K per 
> Inode.  A randomly selected Debian workstation with a lot of packages 
> installed has for it's root filesystem 9.1G and 191111 Inodes for an average 
> of 49K.
>
> As it seems quite unlikely that any non-root filesystem is going to have a 
> smaller average Inode space usage than the root filesystem (I had expected 
> Maildir to be the pathological case of small files) it seems quite safe to 
> make the default be -i 49152 for non-root filesystems and be -i 32768 for root 
> filesystems.
>
> Finally using ext4 features either through "mke2fs -t ext4" or "mke4fs" will 
> give you better fsck performance.  Are we doing ext4 by default nowadays?

The feature relevant here is that the filesystem knows how many inodes
have been used ever and only needs to scan those inodes. So if you only
ever used the first 10000 inodes out of 10000000 that is a huge time
saver.

> As an aside "mke2fs -t ext4" includes huge_file, dir_nlink, and extra_isize 
> while mke4fs doesn't.  This difference seems wrong to me.

Urgs. +1.

MfG
        Goswin