Phil Schwan
2006-May-19 07:36 UTC
[Lustre-discuss] Questions about reiserfs and meta date requirements
On Mon, 2004-04-26 at 03:51, David S. wrote:> > where the devices "/dev/sda1" and "/dev/sdb1" are in the SAN. However, > I haven''t been able to make the above example work with "--fstype ext3" > replaced by "--fstype reiserfs". In that latter case, the "lconf" command > prints the message > > OSD: ost1 ost1_UUID obdfilter /dev/sdb1 0 reiserfs no 0 0 > > and then hangs - no Lustre file system is mounted at "/mnt/lustre". > > So is it the case that Lustre won''t work with a reiserfs "back-end", or > am I doing something wrong?That is correct, ext3 is the only supported file system backend. Given the vast improvements to ext3 over the last few years, we have not found any need, or any customer demand, to support other file systems.> if I have X gigabytes > of data served from my OSTs, how much space should I commit to the meta- > data server supporting them? I''d be gratetul for any guidance.It depends on your own metadata requirements, and is completely unrelated to the amount of OST storage. We use 1% of the disk for the ext3 journal (up to a maximum of 400MB). If you add to that ~4k for each file that you plan to create, that should be plenty (ie, for 10 million files, you should create an MDS of at least 40.4 GB). Thanks-- -Phil
David S.
2006-May-19 07:36 UTC
[Lustre-discuss] Questions about reiserfs and meta date requirements
I''m configuring Lustre 1,0.4 on a dual Zeon 2.4.24 system, patched with with the "vanilla" 2.4.24 patch series from the Lustre distribution, and a fibre channel storage array - I can provide more details on the system and hardware configuration if they''re relevant. Following the example in the Lustre "How-To", section 1.4.2 "Using Supplied Configuration Tools", I''ve been able to get Lustre working with the client, MDS, and OST on a single node with the commands: lmc -o local.xml --add node --node localhost lmc -m local.xml --add net --node localhost --nid localhost --nettype tcp mc -m local.xml --add mds --node localhost --mds mds1 --fstype ext3 --dev /dev/sda1 lmc -m local.xml --add ost --node localhost --ost ost1 --fstype ext3 --dev /dev/sdb1 lmc -m local.xml --add mtpt --node localhost --path /mnt/lustre --mds mds1 --ost ost1 lconf --reformat --gdb local.xml where the devices "/dev/sda1" and "/dev/sdb1" are in the SAN. However, I haven''t been able to make the above example work with "--fstype ext3" replaced by "--fstype reiserfs". In that latter case, the "lconf" command prints the message OSD: ost1 ost1_UUID obdfilter /dev/sdb1 0 reiserfs no 0 0 and then hangs - no Lustre file system is mounted at "/mnt/lustre". So is it the case that Lustre won''t work with a reiserfs "back-end", or am I doing something wrong? My other question concerns meta data requirements: if I have X gigabytes of data served from my OSTs, how much space should I commit to the meta- data server supporting them? I''d be gratetul for any guidance. Thanks in advance, David S. dgs@gs.washington.edu