Hi! I set up some test-environment to gain further insight into the corruptions. This took some time, since I only have one suitable system which at the same time I use as a mailserver etc. First I recreated an almost exact setup of the previous system to see if the corruptions still occur (never underestimate Murphy's Law). This setup looks like this: 2 120gb HDs, each with a 90GB partition for the tests, both partitions combined to a degraded raid5 array (sorry, no other harddrives available to complete the array). Dm-crypt over the md-device, lvm over the dm-crypt device. I started with two large logical volumes and therefore large filesystem, but was now able to reproduce the problem with a smaller (10GB) fs. I'm currently trying 3GB. I'm using reiserfs at the moment, but as was already stated, other filesystems suffer similar corruptions. Other things I'll test after that: - No LVM - Different cyphers - No dm-crypt - Different kernels And the things suggested on this list: - Replace dm-crypt with dm-linear (ie standard logical volume) - Replace md raid5 with md linear - Latest upstream git & -mm kernel (to pick up recent md patches) - Readahead disabled Easiest way for me to cause the corruptions: 1. create a 1GB file containing random noise 2. while [ 1==1 ]; do cat 1GB_NOISE > 1GB_NOISE_2; done After 30 minutes to 3 hours the kernel log will be filled with fs-warnings. On the same harddrives I have a raid1 with dm-crypt but without LVM running perfectly fine. Other details: Ubuntu 6.06 2.6.15-server kernel with the dm-crypt patch applied, that was posted here a few days ago. Servcies running while testing (the system is still in use, sorry): MySQL Apache Postfix Cyrus-Imap Backuppc several system-maintenance related cronjobs If you think it's neccessary I'll disable some of the servcies for a few days. And if I can cough up some money, I'll buy the needed hd to complete the raid5 array (Actually I have a suitable hd, but it contains a backup, which seems neccessary, considering the fs corruptions). Kevin --------------------------------------------------------------------- - http://www.saout.de/misc/dm-crypt/ To unsubscribe, e-mail: dm-crypt-unsubscribe@xxxxxxxx For additional commands, e-mail: dm-crypt-help@xxxxxxxx
