RSS usage higher on newer fedora releases higher for user-applications compared to 3.11.10-301.fc20.x86_64 release, what might be the reason?
So, We were trying to move to Fedora 33 from Centos7 for our jvm and then we found the memory consumption to be higher than Centos7.
we did the following:
Step :1 Narrow the problem with a simple c-code
#include<unistd.h>
int main()
{ pause(); return 0; }
Running this c-program on Fedora 33 yields 660 KB . While on centos7, it only takes 348 KB,
Step 2:
Narrow down the kernel versions .
So, we tried testing different fedora VMs that we had and one of the VMs was running fedora 20. We were able to boot
1. kernel-3.11.10-301.fc20.x86_64
2. kernel-3.17.4-200.fc20.x86_64
With kernel-3.11, pmap tells the RSS was only 388KB . Screenshot of pmap here: (https://ibb.co/D5zTXvs)
while on kernel-3.17 its 920KB (https://ibb.co/vLByKkD) .
From the pmap mappings, we see both the binaries are using the same glibc and ld.so etc. So, this seems like nothing has changed on the user-land while the RSS measurement has more than doubled.
Could somehow have an idea on what else could explain the reason for his memory increase? This seems like something has changed from the kernel side in terms of how RSS was measured or the defaults have changed?
Regards, -Anand
On Tue, Jul 20, 2021 at 08:23:33AM -0000, Anand Babu wrote:
So, We were trying to move to Fedora 33 from Centos7 for our jvm and then we found the memory consumption to be higher than Centos7.
Did the suggestion you got when you posted this on the CentOS list help?
https://lists.centos.org/pipermail/centos/2021-July/354153.html
Thanks for taking the time to comment about the interest in our problem.
I had "digest mode" setup for the email preference and i could not reply to that particular message to the whole group(though i replied to the "digest email" and i could not see that on the mailing list .
I responded to the person(Leon) who made that helpful suggestion directly that it dint change the pmap output and i could still see the original problem.
Now, thanks to your question, here is the full output of trying that suggestion:
1. Before implementing the suggestion:
* cat /sys/kernel/mm/transparent_hugepage/enabledalways [madvise] never*
2. Implemented the suggestion and verified as below:
*[root@fedora-20-64 ~]# cat /sys/kernel/mm/transparent_hugepage/enabledalways madvise [never]*
3. Still, i see the following pmap output:
[root@fedora-20-64 ~]# * pmap -X $(pidof test)1577: ./test* Address Perm Offset Device Inode Size *Rss* Pss Referenced Anonymous Swap Locked Mapping 00400000 r-xp 00000000 00:21 88738 4 4 4 4 0 0 0 test 00600000 r--p 00000000 00:21 88738 4 4 4 4 4 0 0 test 00601000 rw-p 00001000 00:21 88738 4 4 4 4 4 0 0 test 7f3fe3f88000 r-xp 00000000 fd:00 1751 1744 796 12 796 0 0 0 libc-2.18.so 7f3fe413c000 ---p 001b4000 fd:00 1751 2044 0 0 0 0 0 0 libc-2.18.so 7f3fe433b000 r--p 001b3000 fd:00 1751 16 16 16 16 16 0 0 libc-2.18.so 7f3fe433f000 rw-p 001b7000 fd:00 1751 8 8 8 8 8 0 0 libc-2.18.so 7f3fe4341000 rw-p 00000000 00:00 0 20 12 12 12 12 0 0 7f3fe4346000 r-xp 00000000 fd:00 1257 128 128 1 128 0 0 0 ld-2.18.so 7f3fe454e000 rw-p 00000000 00:00 0 12 12 12 12 12 0 0 7f3fe4564000 rw-p 00000000 00:00 0 4 4 4 4 4 0 0 7f3fe4565000 r--p 0001f000 fd:00 1257 4 4 4 4 4 0 0 ld-2.18.so 7f3fe4566000 rw-p 00020000 fd:00 1257 4 4 4 4 4 0 0 ld-2.18.so 7f3fe4567000 rw-p 00000000 00:00 0 4 4 4 4 4 0 0 7fffeded0000 rw-p 00000000 00:00 0 136 8 8 8 8 0 0 [stack] 7fffedf42000 r--p 00000000 00:00 0 8 0 0 0 0 0 0 [vvar] 7fffedf44000 r-xp 00000000 00:00 0 8 4 0 4 0 0 0 [vdso] ffffffffff600000 r-xp 00000000 00:00 0 4 0 0 0 0 0 0 [vsyscall] ==== ==== === ========== ========= ==== ====== 4156 *1012* 97 1012 80 0 0 KB
Enabling that option seemed to have increased the RSS output by about 80KB to 1012(though we do see some noise across different runs to be honest ).
We will be glad to hear any further inputs on this issue.
Regards, -Anand
On Tue, Jul 20, 2021 at 3:45 PM Jonathan Billings billings@negate.org wrote:
On Tue, Jul 20, 2021 at 08:23:33AM -0000, Anand Babu wrote:
So, We were trying to move to Fedora 33 from Centos7 for our jvm and
then we found the memory consumption to be higher than Centos7.
Did the suggestion you got when you posted this on the CentOS list help?
https://lists.centos.org/pipermail/centos/2021-July/354153.html
-- Jonathan Billings billings@negate.org _______________________________________________ users mailing list -- users@lists.fedoraproject.org To unsubscribe send an email to users-leave@lists.fedoraproject.org Fedora Code of Conduct: https://docs.fedoraproject.org/en-US/project/code-of-conduct/ List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines List Archives: https://lists.fedoraproject.org/archives/list/users@lists.fedoraproject.org Do not reply to spam on the list, report it: https://pagure.io/fedora-infrastructure
On Tue, 2021-07-20 at 16:23 +0200, Anand Babu wrote:
I had "digest mode" setup for the email preference and i could not reply to that particular message to the whole group(though i replied to the "digest email" and i could not see that on the mailing list .
Just as an aside, most email clients support replying to a specific message within a digest (rather than to the whole digest), e.g. in Evolution right-click on the message to select the option.
On the whole however, digests are a relic of the days of UUCP mail and are better avoided. One list I moderate disabled them several years ago and we've had no complaints.
poc
-
Anand Babu -
Jonathan Billings -
Patrick O'Callaghan