Very Computer

This did not happen when I was using pl7A and earlier.

I know that the file table is rather small, so it *could* be possible that
it gets really filled, but otherwise this did not happen with the same
mail- and news-soft and pl7A. Also the machine is not loaded at all when
it polls: uucico, uuxqt, rmail and rnews are the only processes running,
so I can't imagine they are using up all filetable entries.

So two questions arise:

   1) Is there really a bug with the handling of the filetable in the kernel,
      or am I just discovering Murphey's law?

   2) Which is the right way to make the filetable larger?
      Is it sufficient just to alter the value of NR_FILE in linux/fs.h
      or do I have to make changes in other places too (i.e. NR_INODE)?.
      Also, will there be problems with applications already compiled when
      I change this value, or do I have to recompile much stuff?

Setup: .99pl9, libc-4.3.2, smail, inn-1.4, taylor-uucp-1.4

JanO
--


6000 Frankfurt 1  | phone: ++49-69-552167             | (_)<(_)

I am also seing this EXACT same behavior. Never got it in releases
pl7a and earlier.

Is there a fix? The problem pretty much renders everything unusable.

Keith Barrett

--
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Linux = Likable Unix :-)               |
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

: I am also seing this EXACT same behavior. Never got it in releases
: pl7a and earlier.

: Is there a fix? The problem pretty much renders everything unusable.

Sure is. Just increase the value of NR_FILE, and make sure that NR_INODE is
bigger than it.

And now that I'm in a mood for pontificating about tweaking code.  Hey,
you've got the source to the kernel. Fiddle with it, and see what happens.
Don't be afraid to change things if you want. Remember, it's *your* unix
system. You can do whatever you want with it.

(Just make sure you have a book disk, though :-)

ciao,
Nick
--



| UCC, Cork, Ireland.        | Disclaimer: Opinions - they're mine, all mine! |

-Joel

--
-----------------------------------------------------------------------------
|_|~~ Germany, Europe. 1943.    "The diameter of the bomb was 30 centimeters,
__|~| 16 Million DEAD.           and the diameter of its destruction, about 7
                                meters, and in it four killed and 11 wounded.
 cnc  Bosnia, Europe. 1993.     And around these, in a larger circle of  pain
 cnc  HOW MANY MORE?          and time,  are scattered two  hospitals and one
                          cemetery.   But the young woman who was  buried  in
                    the place from where she came, at a distance of more than
             than 100 kilometers, enlarges the circle considerably.   And the
      lonely man who is mourning her death in a distant  country incorporates
into the circle the whole world.  And I won't speak of the cry of the orphans
that reaches God's chair and from there makes the circle endless and godless."
-----------------------------------------------------------------------------

This is a solution, have you actually "tried" it?? Tunable kernel features
are there for you to modify if you need a bigger system, use 'em.

James

Quote:>         ...and, the only difference between Linux and the Commercial
> Unix packages is that you have to *search* for the Tunables, but they
> are still there.  

>         Suggestion to Linus:  create a simple list of what Tunables
> their are, and which file to find them in?  I know about tasks.h
> and fs.h...are there any others?

> thanks...

> marc

P.S. How many machine's need the #define REALLY_SLOW_IO? On mine (Everex
Tempo 486/33DX), I've been able to change the #define's to #undef's in
floppy.c, keyboard.c, and hd.c, without any problem. I wonder if there
are any others lurking around I can change. I wonder if most machines
are designed well enough to allow this, so maybe it could be the default
or configurable?

Quote:>         Suggestion to Linus:  create a simple list of what Tunables
> their are, and which file to find them in?  I know about tasks.h
> and fs.h...are there any others?

A list of the tunable parameters, their include file, and comments
(such as range limitations or co-dependicies on other parameters) would be
MOST helpful, and may cut down on some of the newsgroup traffic.

I'd like to suggest that this be created, and perhaps added as a new FAQ
section.

Keith
--
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Linux = Likable Unix :-)               |
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

MF> the solution is to use a dynamic list (right term?) and from
MF> what I've been told, Linus is/has this working, but isn't
MF> prepared to release it without further testing.

Yeah, I hacked, er, wrote it, Linus has it. He is, quite rightly, holding on
to it while all the other stuff dies down a bit :-).

  There does seem to be a race condition in it somewhere. Under *extreme*
thrashing (linking, unlinking, opening, closing etc.) I can make things die
with permission denied errors on the shared libraries.

  I also notice a possible memory leak with it when searching for the above
but since it would get triggered once every thousandth blue moon I may have
been imagining it.

  No, I don't think a public unofficial patch is likely... :-).

                                Mike