enrico.sirola@gmail.com
2006-May-19 07:36 UTC
[Lustre-discuss] lustre and scientific linux 4.1
Hi, I''m having problems while trying to use lustre on SL4.1 for amd64. The situation is as follows: the system is running SL4.1 (x86_64 edition), which basically is a RHEL clone. The installed kernel is 2.6.9-22; uname -a reports the following: enri@sirola01 % uname -a Linux sirola01.statpro-it.lan 2.6.9-22.0.1.EL #1 Thu Oct 27 11:29:04 CDT 2005 x86_64 x86_64 x86_64 GNU/Linux the system is in sync with the scientific linux "bugfix", "contrib", "errata" and "base" yum repositories. When I install the 1.4.5 lustre kernels (I used the rpms for the lustre site for versions 2.6.9-5 for RHEL4) the system isn''t unable to bootstrap: the kernel is unable to find the volume group disk1. The lvm volumes are as follows: lvm> lvs LV VG Attr LSize Origin Snap% Move Copy% home disk1 -wi-ao 7,81G opt disk1 -wi-ao 512,00M root disk1 -wi-ao 1,00G swap disk1 -wi-ao 2,00G tmp disk1 -wi-ao 1,00G usr disk1 -wi-ao 7,81G usr_local disk1 -wi-ao 1,00G var disk1 -wi-ao 1,00G while /boot resides in a primary partition. Are there problems with root partitions residing on logical volumes? I''m asking it because on my laptop everything seems to work and it has the root directory mounted from a primary partition. Please let me know if you need additional info. thanks in advance and bye, enrico -- Enrico Sirola <enrico.sirola@statpro.com>
enrico.sirola@gmail.com
2006-May-19 07:36 UTC
[Lustre-discuss] lustre and scientific linux 4.1
>>>>> "Andreas" == Andreas Dilger <adilger@clusterfs.com> writes:[...] Andreas> The kernels we use are built with the same config options Andreas> as the RHEL kernels. This puts almost everything into Andreas> modules, and RHEL/FC kernels all use init ramdisks to Andreas> load modules (including lvm + ext3). I had previously Andreas> set up a system with root on LVM, and this worked fine Andreas> until one day I updated the kernel (not Lustre related) Andreas> and the initrd didn''t work properly and the system Andreas> wouldn''t boot anymore. I now always put at least boot + Andreas> root into a regular partition to avoid this problem. Andreas> It may be that SL has LVM compiled into the kernel. ok, today I''ll try a FS layout with / + /boot on the same primary partition. >> while /boot resides in a primary partition. Are there problems >> with root partitions residing on logical volumes? Andreas> I suspect you need to fix up your initrd, but that is Andreas> beyond the scope of this mailing list. Ok, thanks a lot, e. -- Enrico Sirola <enrico.sirola@statpro.com>
On Nov 21, 2005 15:17 +0100, enrico.sirola@gmail.com wrote:> I''m having problems while trying to use lustre on SL4.1 for amd64. The > situation is as follows: the system is running SL4.1 (x86_64 edition), > which basically is a RHEL clone. The installed kernel is 2.6.9-22; > uname -a reports the following: > > enri@sirola01 % uname -a > Linux sirola01.statpro-it.lan 2.6.9-22.0.1.EL #1 Thu Oct 27 11:29:04 CDT 2005 x86_64 x86_64 x86_64 GNU/Linux > > the system is in sync with the scientific linux "bugfix", "contrib", > "errata" and "base" yum repositories. > When I install the 1.4.5 lustre kernels (I used the rpms for the > lustre site for versions 2.6.9-5 for RHEL4) the system isn''t unable to > bootstrap: the kernel is unable to find the volume group disk1.The kernels we use are built with the same config options as the RHEL kernels. This puts almost everything into modules, and RHEL/FC kernels all use init ramdisks to load modules (including lvm + ext3). I had previously set up a system with root on LVM, and this worked fine until one day I updated the kernel (not Lustre related) and the initrd didn''t work properly and the system wouldn''t boot anymore. I now always put at least boot + root into a regular partition to avoid this problem. It may be that SL has LVM compiled into the kernel.> while /boot resides in a primary partition. Are there problems with > root partitions residing on logical volumes?I suspect you need to fix up your initrd, but that is beyond the scope of this mailing list. Cheers, Andreas -- Andreas Dilger Principal Software Engineer Cluster File Systems, Inc.