There are three core parts of XtreemFS system:
- osd - object storage data (actual data storage)
- mrc - meta data and replica catalog server (headers storage)
- dir - directory (directory service)
(Note: dir service is similar to 'mon' service in Ceph
Debian Package Setup
It is recommended to have the Linux kernel 2.6.36 RC 7 (this is the October 2010 kernel).
Download and Install
We are using Ubuntu
and the build process is described in detail at external: https://www.xtreemfs.org/quickstart.php
Download the following packages from external: https://www.xtreemfs.org/download.php?t=debs
Install the packages
$ dpkg -i xtreemfs-client-1.2.3.deb xtreemfs-server-1.2.3.deb xtreemfs-backend-1.2.3.deb
on all the machines (Note. You can only install xtreemfs-server on server machines and xtreemfs-client on client machines, but all machines need xtreemfs-backend and probably xtreemfs-tools).
For details see the single machine install & setup. external https://www.xtreemfs.org/quickstart.php?t=d
General note: very important to follow step by step;
Create a directory for OSDs at
- s1 - osd0 (s1)
- s2 - osd1 (s2)
- s3 - mrc (s3)
- s4 - dir (s4)
plus one additional node for client (in this case, we will use s4). All OSD, MRC and DIR should have the configuration file /etc/xos/xtreemfs/*.properties
- - dirconfig.properties (dir service)
- - mrcconfig.properties (mrc service)
- - osdconfig.properties (osd service)
Since the setup is configured using *.properties, except for the dir service (s4), both osd and mrc must have their respective property file changed to point to the location (IP or hostname) of dir,
- For MRC (s3), edit /etc/xos/xreemfs/mrcconfig.properties, and change the default line dir_service.host = localhost to dir_service.host = s4
- For OSD1 and OSD2 (s1 and s2), edit /etc/xos/xreemfs/osdconfig.properties, and change the line dir_service.host = localhost to dir_service.host = s4
Also, you need to make sure you are mounting the OSDs in the correct location
- For OSD1 and OSD2, make sure to mount the device to "/var/lib/xtreemfs/objs/" as indicated in the line, object_dir = /var/lib/xtreemfs/objs/ It is recommended to mount ext3/ext4 file system.
Now, start up the servers (dir, mrc, osd). We recommned this order
* /etc/init.d/xtreemfs-dir start
* /etc/init.d/xtreemfs-mrc start
* /etc/init.d/xtreemfs-osd start
Note that XtreemFS does not output any errors, so make sure to check the log messages
ps -Af | grep xtreemfs
to check that the services are running.
See external https://www.xtreemfs.org/quickstart.php
from more details.
This part involves two steps:
- Create a volume on the MRC-- e.g., Normal, Stripe (RAID0), Replica mode (Note. Volume location is always at mrc (s3).)
- Mount the created volume (note - mount location is always at * dir (s4)*
Example (Basic FS)
- Create FS called 'test' (you must create it to the MRC location)
$ sudo mkfs.xtreemfs s3/test
- Mount the 'test' (you must mount it to the DIR location, and make sure to create empty directory at /media)
$ sudo mount.xtreemfs s4/test /media/test
Example (RAID-0 for osds on s1 and s2)
$ sudo mkfs.xtreemfs -p RAID0 -w 2 s3/raid
Useful additional parameters for mounting
- -p accepts RAID0 only so far (1.2.x)
- -w is the 'width' i.e., how many OSDs are to be used for the striping (in round-robin fashion), in this case we have 2 OSDs.
- -s is the stripe-size in KB, by default it is 128K (e.g., -s 64 to set to 64KB)
$ sudo mount.xtreemfs -o allow_other s4/raid /media/raid
$ sudo mount.xtreemfs --cache-data --metadata-cache -o direct_io -o allow_other s4/raid /media/raid
- --cache-data: enable client-side data caching
- --cache-metadata: enable client-side metadata caching
- -o (FUSE option): direct_io, allow_other, big_writes
For more information see
Check the current OSD configuration
Individual files can have its own file striping methods, e.g.,
$ sudo xtfs_sp --get /media/test
stripe-size (kB): 128
Each OSD/MRC/DIR must be uniquely identifiable by others.
For OSDs (in our case s1 and s2), each OSDs must have UUID.