inodes: df -i vs /proc/sys/fs/inode-nr

[Marco Shaw]

(I've tried, but failed to find this answer elsewhere, or at least a
definite answer, without trying to look at the source code.)

Is there anything online that contrasts what "df -i" reports and the
values provided by /proc/sys/fs/inode-nr?  Is there a connection?  I
have a decent grasp on VFS-related "inodes" on disk, but inode-nr
reports "in-memory inodes"(?), and I don't understand what that means.

For a particular system, for example, df -i reports 1% to 4% inode
utilization for different FSs, while /proc/sys/fs/inode-nr reports 70%
(even 100% at some point) used (see Note 1 below).

For example, what happens when the inode-nr reports 100% used on a
system?  I opened a support case with my vendor, and at one point an
engineer indicates that "if inode-nr reports 100% utilization, there
could be all kinds of problems", but then another engineer goes on to
indicate something along the lines that "with kernel 2.6, this is no
longer a problem and the kernel will dynamically adjust the maximum".

So, should I monitor /proc/sys/fs/inode-nr?  Should I only monitoring
it on 2.4.x because it is "meaningless" (or at least not a cause of
concern) on 2.6.x?

Marco

Note 1: inode_used_percent = 100*(inode_alloc - inode_free)/inode_alloc
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/