From Bugzilla Helper:
User-Agent: Mozilla/5.0 (Windows; U; Windows NT 5.1; rv:1.7.3)
Description of problem:
Error affects GeoViz, an application from SIS.
When memory is allocated and freed, this is not available, until 15-30
minutes later, even if the application exit.
Apparently, this is similar to
Version-Release number of selected component (if applicable):
Steps to Reproduce:
1.Run a program that allocates more than 50% of available RAM + Swap
3.Immediately rerun the above program
Actual Results: Swapping or memory allocation error
Expected Results: no memory allocation error
First of all, the normal system behavior is to hold on pagecache pages
that a process accessed rather than freeing upon exit. This works
this way so that when another process accesses the same file pages
they will already be in memory and won't have to be re-read. What
this means to users and/or casual observers is that before you start
running an application there is more free memory than when the
Having said that, the kernel is responsible for reclaiming pages when
a memory deficit occurs. We have made several changes to the kernel
in RHEL3-U3 and RHEL3-U4 that expedite reclaiming pagecache pages
rather than swapping anonymous memory pages when the system runs out
You really need to run the very latest RHEL3 update kernel in order to
have the system reclaim pagecache pages instead of swapping.
Can you test with the latest RHEL3-U4 kernel? We can provide you with
the latest RHEL 3 U4 kernel (2.4.21-22.EL) once it emerges from QA
(later today or tomorrow). The current RHEL 3 U4 beta ISOs, which are
posted to Red Hat Network (RHN) contain the 2.4.21-21.EL DO NOT
contain the fix that you'll need. We'll send mail and also post the
location of the -22.EL kernel here.
Arun, please try the latest RHEL3-U4 kernel and see if that fixes the
problem you are seeing. Like I said above on 10/27 its normal
behavior for the Linux kernel to hold onto filesystem cache pages so
you cant expect the the system to have the exact number of free pages
when the program ends as it did when it first started. However if
this is causing memory allocation failures especially at the user
level there is certainly a bug somewhere.
Please try out the latest RHEL3-U4 kernel located here:
Arun, we are waiting on you to confirm whether this problem has been
fixed. Please download the appropriate RPM from the URL specified in
comment #4 and update this bug report. Thanks in advance.
(The kernel version should be 2.4.21-27.EL.)
Actually, my team attempted to use this and the feedback is enclosed.
This link only points to the single processor version. We also need the "smp"
version of this. I tried going up one level and adding "smp" to the link and
neither gets me to a place where I can find the "smp" version. There is no sense
testing this unless we can also verify that it works in multiple processor mode.
Can you contact the people at RedHat and get a link to the "smp" version?
Arun, its there. This link points kernel rpms for every kernel configuration
for every architecture we build:
Specifically this is the exact link to the x86_64 smp kernel:
We will give it a try right away.
Hello, Arun. Any confirmation that Larry's test kernel resolved this problem?
Also, there have been two security errata issued since U4, so if you're going
to begin your testing from the beginning, it would be better to use the latest
2.4.21-27.0.2.EL package from RHN (which was released last night).
I can confirm based on our testing that we are no longer seeing this problem
Thanks for all your help, it is much appreciated.
A fix for this problem was committed to the RHEL3 U4 patch pool
on 23-Sep-2004 (in kernel version 2.4.21-20.11.EL), and this was
later followed up by a 2nd fix committed on 18-Oct-2004 (in kernel
Thanks for the confirmation, Arun.
An errata has been issued which should help the problem
described in this bug report. This report is therefore being
closed with a resolution of ERRATA. For more information
on the solution and/or where to find the updated files,
please follow the link below. You may reopen this bug report
if the solution does not work for you.