On Jul 29, 2008, at 6:01 AM, Andrew Doran wrote:
- Adopting a producer-consumer model where journal writeback is handled by akthread would remove the need for the rwlock. It could be done bymaintaining transaction sequence numbers. For example, each in-core inodecould contain a record of the last transaction made and the kthread aglobal sequence number indicating the last transaction that is known to bein stable storage.
Yah, that sounds like a good solution. -- thorpej