implement and test write-anywhere model

A major performance overhead that ddsnap adds is the copyout before
overwriting the old origin data. To reduce the overhead of copyout, we can
switch to the write-anywhere model. I.e., we don't overwrite the old data
on the origin device. Instead, new data is written to a newly allocated
chunk with btree being properly updated. To reduce data fragmentation, a
background process will copy the new data to the old chunks periodically.

Original issue reported on code.google.com by jiahotc...@gmail.com on 9 Apr 2008 at 5:21

junneyang / zumastor

implement and test write-anywhere model #108