Re: Could not get a free page...

Dale Scheetz (dwarf@polaris.net)
Fri, 30 May 1997 16:10:01 -0400 (EDT)


On Thu, 29 May 1997, Pavel Machek wrote:

> hi!
>
> I just wanted to make sure:
>
> Is "Could not get a free page" normal condition, or it just should not
> happen? I always thought that system is expected to work even without
> free pages and that Could not get a free page is something just normal
> which will happen everytime...
>
This is a REAL problem with the 2.0.30 kernel and to a much more limited
extent the 2.0.29 kernel as well. The problem has gotten bad enough with
the 2.0.30 kernel that even simple installation operations become
impossible under what are normaly considered useful machines.

We have two almost identical 486 machines with 8 meg of memory being used
for installation testing. Both of these machines will lock up during
mke2fs on partitions larger than 600 meg. 800 meg partitions will lock up
every time. In both cases the kernel has a swap partition of 80 meg
available (no swaping to speak of). By upping the priorities for freepage
in /proc the threshold for failure rises a bit.

I also have a report of a machine with 16 meg of real memory locking up
while creating a file system on a 2.1 gig partition. You can do the math,
but this doesn't look linear to me.

As I understand it, the activity that triggers this failure centers around
buffer allocation. Any process that allocats a large number of buffers in
a short period of time can overrun the paging process and lock out access
to swap. This problem exists in the 2.0.29 kernel, but in a more quiescent
form that doesn't rear its ugly head except under extreme load conditions.
If the problems gets any worse in 2.0.31 the kernel will be useless for
even the most mundane file management.

Is anyone working on fixing this problem?

Thanks,

Dwarf

-- 
_-_-_-_-_-_-                                          _-_-_-_-_-_-_-

aka Dale Scheetz Phone: 1 (904) 656-9769 Flexible Software 11000 McCrackin Road e-mail: dwarf@polaris.net Tallahassee, FL 32308

_-_-_-_-_-_- If you don't see what you want, just ask _-_-_-_-_-_-_-