On Tue, 20 Jan 2009, Stas Oskin wrote: > Can someone explain in layman terms what is the meaning under the term " > shared block storage"? As far as I understand it's a shared storage device > (SAN, plain hard-disk, etc...), which provides write/read access to multiple > servers/clients, and coordinates all the access operations (i.e. locking > files, keeping files shadow copies, i.e.). I will try. Shared block storage is a block device like /dev/sda accessible on more than one server at once. This may be a plain hard-disk or a hardware RAID. Coordination among the servers is handled on the file system level above that - for example by OCFS or GFS. Using a normal file system like Ext3 is not possible for read/write operations to the same shared block storage since the servers would get very confused. > Also, can someone explain where actually the shared block storage appears in > Lustre? From what I read, the data is stored on separate storage nodes, and > not on shared disk. As far as I know; Lustre in itself do not handle redundancy or fault tolerance. In order to get redundancy for Lustre (unlike GlusterFS or Ceph) that must be handled by some kind of shared storage. And Lustre needs the data to be stored on a block device. Using shared block storage to keep the data on (usually a hardware RAID on a SAN) makes it to move the service from one physical server to another. Example: normally one server is taking care of service A and another server is taking care of service B. The first server is also passively taking care of service B and the other server passively taking care of service A. When the first server goes down the other server is taking over service A, which means both A and B are located on the same server. Regards, Jerker Nyberg.
