| 摘要 |
In a disk array system for RAID (level 5) to improve a process performance by dispersing data, the overhead during data writing operation is reduced. Even when write data are used to rewrite data #1 to data #4, which have already been written at address SADR 2, SADR 2, SADR1 and SADR 3 within drives as data belonging to mutually different parity groups, these data are considered as new write data, and then these new write data are written in parallel into, for instance, an empty region in the drive at the address SADR 4. The updated old data or the updated old parity data are not read out. The invalidation flags for the updated old data are registered into an address conversion table. The subsequent data readout operation is carried out from a newly written region. When all of data within the original parity group are invalidated, the region holding this group is used as an empty region. The valid data within the parity group, which have been partially invalidated, are replaced at a proper timing. In other words, these valid data are acquired to produce a new parity data group which will then be stored into the empty region. In the above operation, it is assumed that a length of data transferred from a host unit is constant, and data belonging to one parity data group has this constant length. |
