> [Alan, Bill and Stephen, please read on. I think we may have uncovered
> a bug here (I'm not familiar with the code though, so I may have
> skipped over something).]
I'll be happy to look into this, but some further information would be
helpful.
Under normal conditions on the system in question, what are the values
for file-nr, file-max, inode-nr and inode-max? (These are in
/proc/sys/fs/{name})
Then when the system reports "too many open files", what are the
corresponding values?
Finally, are there any messages in the syslog to indicate aloocation
failures, etc?
Regards,
Bill
>
>
> On Thu, 8 Oct 1998, Matthew Hawkins wrote:
>
> > G'day smiley happy people :)
>
> Hmm, that REMinds me of some song...
>
> > I have this problem with 2.1.124, and both .125 pre-patches. It
> > seems that a lot of programs are refusing to function as they
> > complain about "too many open files" and so forth. Now I know this
> > is rubbish as the same programs happen to function fine with 2.0.35,
> > and 2.1.119. I can reproduce it with Debian's apt, Netscape
> > Navigator 4.06, and XF86_S3.
>
> > Now what I think seems to be the problem is the default setting of
> > /proc/sys/fs/inode-max. The documentation from Rik says that this
> > setting should be 3-4 times greater than file-max. file-max is of
> > course set to 4096 by default, and to my surprise, so is inode-max.
> > Increasing inode-max to 16384 seems to get rid of the "too many open
> > files" problem.
>
> > What's the story here?
>
> The inode cache was using far too much memory on a lot
> of small machines, so the number of inodes cached was
> dynamically set on boot dependant on the amount of memory
> someone has.
>
> The inode cache does the following things:
> - keep track of open files/sockets/pipes/fifos/other stuff
> - cache inodes of files on disk as referenced by the dcache
>
> Since the inode cache functions as a slave cache to
> the file cache, I guess that your machine already used
> all inode cache entries to cache files cached by the
> dcache, leaving:
> - no free entries
> - no possibility of freeing an entry, since that also
> has to be done at the dcache level
>
> You seem to have uncovered a bug, raising the number
> of cached inodes is a workaround, but you can expect
> a fix real soon (if I identified it correctly, I'm
> CC:ing some folks in order to check if my suspicion <sp?>
> is right)...
>
> cheers,
>
> Rik.
> +-------------------------------------------------------------------+
> | Linux memory management tour guide. H.H.vanRiel@phys.uu.nl |
> | Scouting Vries cubscout leader. http://www.phys.uu.nl/~riel/ |
> +-------------------------------------------------------------------+
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/