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
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):
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