Linus Torvalds wrote:
> 
> If we're going to do this for a major filesystem, then I'd really just
> rather see this being done generically during 2.5.x, making "u" be a
> pointer only, and having the generic "iput()" just always free the dang
> thing.

Here are the actual sizes, on x86 SMP, from 2.4.5:

nfs_inode_info = 196
ext3_inode_info = 120
ext2_inode_info = 116
efs_inode_info = 104
ufs_inode_info = 100
affs_inode_info = 96
shmem_inode_info = 92
umsdos_inode_info = 68
qnx4_inode_info = 68
coda_inode_info = 68
minix_inode_info = 64
hpfs_inode_info = 64
reiserfs_inode_info = 56
ncp_inode_info = 56
sysv_inode_info = 52
udf_inode_info = 48
socket = 44
ntfs_inode_info = 44
hfs_inode_info = 44
msdos_inode_info = 36
adfs_inode_info = 28
smb_inode_info = 24
usbdev_inode_info = 20
proc_inode_info = 16
iso_inode_info = 16
bfs_inode_info = 12
romfs_inode_info = 8

Tell me what's wrong with this chain of logic:

- Almost all machines use ext2

- Splitting the inode is detrimental for ext2 because there
  are two things to allocate. (There may be some small benefit
  from page-packing with two objects)

- Only a very small number of systems *mainly* use a filesystem
  whose private inode info is significantly smaller than
  ext2.  (ie: we're penalising sysvfs-only machines).

So the rule should be: if your private inode info is larger
than ext2's, you should allocate it separately.
-
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to [EMAIL PROTECTED]

Reply via email to