[Ocfs2-users] OCFS2 and berkeley database files

[Joel Becker]

On Wed, Dec 06, 2006 at 05:39:28PM -0800, Alexei_Roudnev wrote:
> Many sources are saying, that maintaining syncronzied memory (mmap to the
> same area == you must syncronized shared memory between 2 servers) is not a
> trivial task. Oracle is doing it inside RAC luster for some objects only,
> using CSSD.

	It's not trivial, but it is certainly doable.  The memory does
not have to be fully synchronized, nothing in a cache is.  All that
matters is that it _appears_ synchronized.
	The kernel provides mechanisms to know when a memory page is
written to.  OCFS2 must have all other nodes flush their copy.  Then,
when the other nodes want to read that memory page, OCFS2 knows to load
it from the disk.
	It's very similar to sharing page cache data for read(2) and
write(2).  When a writing node modifies some data, the other nodes are
told to forget their old copies.  The only difference is in the kernel
mechanisms (the read and write paths are easy, whereas the page access
paths are more complex).

Joel

-- 

"But all my words come back to me
 In shades of mediocrity.
 Like emptiness in harmony
 I need someone to comfort me."

Joel Becker
Principal Software Developer
Oracle
E-mail: joel.becker at oracle.com
Phone: (650) 506-8127