Having noted that OCFS2 recently landed in Ubuntu Linux, and having for about 9 months used Lustre is a small configuration, I'm thinking of trying OCFS2 instead. But there's some questions not or unclearly answered in the FAQ, so here goes: Can you mmap a file on OCFS2 for reading? Can you execute programs off OCFS2? Can two hosts open the same file O_APPEND, with correct results? Can the OCFS2 volume be on top of an LVM device? MD? How would that work? Can you have 32-bit and 64-bit nodes in the same cluster? Nodes with different page sizes (e.g. Itanium)? Is there a limit on the size of the volume on 64-bit hosts? On 32-bit? How many inodes will I get from e.g. a 4TB volume? Is there a subdirectory limit? Where can I find some comparative benchmarks? How close to wire speed can OCFS2 achieve on a single host? Cluster-wide? How many stats/second can it do? How much net traffic is caused by a given amount of fibrechannel traffic? Can you safely deliver mail on OCFS2? -jwb
Jeffrey W. Baker wrote:>Can you mmap a file on OCFS2 for reading? > >Yes>Can you execute programs off OCFS2? > >Yes>Can two hosts open the same file O_APPEND, with correct results? > >Not tried it personally. But it should work. File a bug if it does not. :)>Can the OCFS2 volume be on top of an LVM device? MD? How would that >work? > >The LVM would have to be cluster aware. While we have not tested this as yet, there is no reason why this should not work.>Can you have 32-bit and 64-bit nodes in the same cluster? Nodes with >different page sizes (e.g. Itanium)? > > >Yes. We support x86, x86-64 and IA64. All can access the same fs concurrently.>Is there a limit on the size of the volume on 64-bit hosts? On 32-bit? > >The limit is 4 billion clusters. So if the cluster size is 1MB, it's 4 Peta bytes.>How many inodes will I get from e.g. a 4TB volume? > >The inodes are allocated from a dynamically growing space. That is, we do not pre-allocate inode space. Each inode is atleast a block.>Is there a subdirectory limit? > >Unless vfs has some limits. no. Not that I am aware of.>Where can I find some comparative benchmarks? How close to wire speed >can OCFS2 achieve on a single host? Cluster-wide? How many >stats/second can it do? How much net traffic is caused by a given >amount of fibrechannel traffic? > >We don't have any official benchmarks as yet.>Can you safely deliver mail on OCFS2? > > >Care to expand.
On Mon, 2005-08-08 at 10:10 -0700, Sunil Mushran wrote:> Jeffrey W. Baker wrote: > >How many inodes will I get from e.g. a 4TB volume? > > > > > The inodes are allocated from a dynamically growing space. That is, > we > do not > pre-allocate inode space. Each inode is atleast a block.On a 4TB volume with 4KB blocks, the maximum number of inodes will be 1 billion?> >Is there a subdirectory limit? > > > > > Unless vfs has some limits. no. Not that I am aware of.I ask this question because ext2/ext3 have a limit of 32,000 subdirectories.> >Can you safely deliver mail on OCFS2? > > > Care to expand.Mail delivery is a kind of acid test of filesystem usefulness. You cannot safely deliver mail without an atomic link(2) or rename(2). For example, you cannot safely deliver mail on NFS. Regards, jwb