SUMMARY idnode problem?

From: John Ciesla (
Date: Fri Apr 23 1993 - 08:33:52 CDT

> Having problems this end, getting out.
> test
> Anybody get this?
I received your message.

> If so, the inode count on one of my workstations is 256/256 when doing
> a pstat -T. Obviously I am running out of inodes. Why is this happening
> and how do I fix it?
> The system is a sparc 1+ with a 370Meg Anthem drive with plenty of room
> on the disk.
First off: Thank you everyone for your responses and test messages.
It appears as though we fixed our problems by viewing some of your headers.

As far as the inode problem.
I received this from:!stern (Hal Stern - NE Area Systems Engineer)

you aren't running out of inodes. if you were, you would see messages
like "/: out of inodes" and the system would hiccup nastily.

the inode count shown by pstat -T is the current usage and high
water mark of the in-core inode table. when you open a file, or
at least fetch its attributes, the inode for that file is loaded
into the in-core cache. this avoids having to do repeated disk
accesses to update modification times and sizes during a long series
of I/O operations. the size of the in-core table is clamped
at the maximum determined by the kernel variable "ninode", which
is scaled based on the "maxusers" variable (see your kernel config file).

but since inodes are non-trivial in size, the kernel only allocates
room for them as needed, rather than in one big chunk at boot time.
so the inode cache grows over time. if you release some inodes
(close some files, for example) and they get dropped from the cache,
the current cache usage might dip below the high water mark. but pstat -T
only shows you the usage and "largest size since boot", not the maximum

now it may be that 256 is the max size and is too small (it's full
at 256 entries as pstat -T shows you). use vmstat -b to look at
the buffer cache hit rates -- the buffer cache is used to move
inodes in/out of the cache. if you see a low hit rate there, it
means you're dragging a lot of inodes around -- crank up maxusers
to make the inode cache bigger.

I also wish to thank the following who were also right on track:

Eddy Eldridge <jvncnet!!eddy>
"James D. Watson" <jvncnet!!JW1675A>
(Manish Bhatia)
Tom Conroy
michael pearlman <>
Sanjay Nadkarni
Wilson Receda Internet :

================================================================================John Ciesla Voice (w/mail): 609-520-5105
Dow Jones & Co., Inc. Fax: 609-520-5089
Engineering Department
P.O. Box 300
Princeton, NJ 08543-0300
Route 1 & Ridge Road
South Brunswick, NJ 08852


This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:07:46 CDT