| 摘要 |
A method for enhancing the performance on non-full stripe writes while closing the RAID5 write hole is disclosed. When a RAID controller receives data to be written to a disk array, the controller stores the data in nonvolatile memory, and opens a write operation. When the current write operation is initiated, the controller reads the old data from the disk array and checks for cached parity that corresponds to the stripe of data that is to be modified. If the parity is cached, the controller reads the cached parity, and if the parity is not cached the controller reads the old parity from the disk array. New parity is then computed. If the parity was cached, the controller will modify a parity log to show an outstanding data write. If the parity was not cached, the controller will open a parity log to show an outstanding data and parity write. The new parity is then cached in non-volatile memory. The controller then determines whether the following write operation is for the same stripe of data. If the following write operation is for the same stripe, the controller writes the new data to disk and opens a parity log for the following write operation to show an outstanding parity write, and terminates the current write operation. If the following write operation is not for the same stripe, the controller writes both the new data and new parity to disk, and terminates the current write operation.
|
