search

nilfs-dev / nilfs2-kmod6

NILFS2 stand-alone kernel module for RHEL 6 clones (this module is no longer maintained; do not use)
https://nilfs.sourceforge.io/
GNU General Public License v2.0
6 stars 0 forks source link

Abysmal performance and WARNING: at mm/page_alloc.c #1

Open aride opened 8 years ago

aride commented 8 years ago

Hello,

I'm trying NILFS2 1.1-4.el6 on OpenVZ kernel 2.6.32-042stab113.11 on an IBM System x3650 M4 with dual CPU, 64 GB RAM and some hardware-RAID1 array of SAS HDDs with an ext4 filesystem on LVM2 over LUKS. The NILFS2 device is just a loop-mounted file in that ext4 filesystem.

The moment I start copying files with rsync from somewhere on the same filesystem into the NILFS device, I start to get tons of repeating warnings starting with the following text (see attached file):

WARNING: at mm/page_alloc.c:2421 __alloc_pages_nodemask+0xaa/0xbb0()

Files get copied anyway, and NILFS works, but the performance is beyond abysmal. For comparison, I tested the same kernel and module with the same setup on a much smaller IBM System x3250 M3 with a single CPU, only 6 GB RAM and two SATA HDDs in software RAID1 (md), with the same NILFS2-loop-ext4-LVM2-LUKS setup and it works quite well (of course the performance is not great but it's several orders of magnitude greater than I get with the beefier server, so I think something's not right with NILFS).

The warning messages are so many that I wonder if the performance degradation is not effectively caused by them. I tried setting kernel.printk to 0 0 0 0 but still couldn't keep them at bay, although maybe that would be some error on my part or some unrelated bug because I don't know why that wouldn't stop them.

The performance degradation for the whole server was serious, even after lowering the rsync I/O priority with ionice.

Attachment: nilfsbug.txt

aride commented 8 years ago

BTW, this was all tested in the Hardware Node, that's what OpenVZ calls the main system, and not inside any OpenVZ containers so I think for all purposes this kernel is the same as CentOS 6/RHEL 6.

aride commented 8 years ago

On second glance, I see the same warnings in the smaller servers too, but they do function with a reasonable performance (writing on NILFS at about 37 Mbps if my quick calculation was correct). So the messages themselves do not seem to be the culprits, just a symptom. I attach the warnings from the small server so as to compare it with the previous one.

Attachment: nilfsbug2.txt