i-node percentages

[Michael De Loe_1]

When we issue the 'df -i' command the total number of i-nodes used, free, in total and percente used is displayed.

Why would the percent used by greater than 100%? Would this indicate a performance issue? And, how best to correct.

Thanks,
Mike

[Alan Riggs]

Perhaps you could post the actual numbers you are seeing. Is the percentage calculaiton accurate? Does the df -i command show more inodes used than total?

As to performance issues, the number of inodes in use does not have a direct relationship to perormance. It certainly might be an indication that your filesystem might be becoming congested which might lead to increased time for directory seeks which can have a performance impact, but this is certainly not one of the first places I look for tuning.

In general, a vxfs file system will automatically allocate new inodes when the current allotment is exhausted (Unless a hard limit was set with mkfs at filesystem creation), so using all available inodes on a vxfs filesystem is not usually a concern. An hfs filesystem, however, has a static inode max which will not allow new files to be created once it is reached. Approaching 100% inode use on an hfs filesystem is an area of concern.

[Anthony deRito · Accepted solution]

df is the generic version block report command and bdf is the Berkeley verison of the block report command. Try and use the bdf -i command and see if you notice accurate figures. I am not sure whats going on with df -i.

Tony

[James R. Ferguson]

Hi:

You must have a very large number of very small files to have exhausted the inode count before filling the actual file system.

If the percentage is near 100 (say 101) I would assume that you are seeing a rounding error in the calculation.

I also assume that your filesytem is journaled (vxfs), in which case inode numbers are unlimited by default. Inodes in this case are created as long as there are free blocks of disk available. See the man page for 'mkfs_vxfs' for more information.

...JRF...

[Michael De Loe_1]

Our application is constantly reading directory contents and acts on the contained files accordingly, so there is a significant amount of disk I/O. When we noticed the percent used to be greater than 100, a red flag was raised.

Thanks,

Actual I-node Numbers:

/home (/dev/vg00/lvol4 ) : 251648 total i-nodes
52269 free i-nodes
199379 used i-nodes
381 % i-nodes used

/apps/vision/spool2 (/dev/vg00/lv_spool2 ) : 2527964 total i-nodes
940846 free i-nodes
1587118 used i-nodes
168 % i-nodes used

[Patrick Wallek]

I would be tempted to call the HP Response Center on this issue. It looks as if the calculation for percent of inodes used is screwed up. You've still got free inodes on both of those lvol's.

[Alan Riggs]

Yes -- it looks like the df command is having trouble with its arithmetic. You are not near exhausting your allocated inodes.

[Michael De Loe_1]

There must be a problem with 'df -i' command. When I use 'bdf -i' the percentage is what you would expect.

Thanks for the help.

[James R. Ferguson]

Michael:

Interesting. Your numbers don't generate the correct percentages! Equally interesting when I looked on one of my servers, df -i reported 77% used for /usr but bdf -i reported (correctly) only 44%. I've never used df, always bdf.

...JRF...

[John Palmer]

I would be interested in what 'bdf -i' reported, I never ever use 'df'.

If possible, I would try unmounting the filesystems and remounting them to see what df and bdf reported then.

VXFS filesystems by default are created with 'unlimited' i-nodes. This means that when you use up all the ones created initially, more can be created dynamically. I wonder if this could have happened and confused 'df'.

The other possibility is that it's a bug in 'df' but I can't find any known problem.