3.5.8 Hardening of data
On modern disk subsystems, all WRITE CCWs actually transfer data into subsystem cache memory before writing the data to the back-end disks. Data is written asynchronously from the cache to the back-end disk. If a subsequent WRITE modifies the same track before it is “hardened” from cache to disk, the previous data is discarded and never written to disk. The proper operation of FDRERASE requires that the program insure that the data is hardened to disk at the end of each pass.
• On IBM and HDS subsystems, a COMMIT CCW is supported. COMMIT will insure that all cached data for a range of track addresses on a given device has been written to the back-end disks before any new I/O is accepted.
• On EMC subsystems, COMMIT is not supported but unique EMC query commands will return the number of writes which are pending in the cache for a given disk device. This can be queried repeatedly at intervals until it reaches zero.
• However, all of these vendors guarantee that write data will eventually be hardened to the back-end disk, even if a power failure occurs.
For ERASE and SECUREERASE operations, FDRERASE issues the appropriate commands to insure that the data is hardened “at the end of each pass”. When multiple passes are made on a disk special techniques are used to improve performance for example:
• The first pass will write on all cylinders of the device, from the highest-number cylinder to cylinder 0.
• A COMMIT will be issued for the top half (highest-numbered cylinders) of the device, since that data has probably already been hardened by the subsystem.
• The next pass starts by writing the top half of the device with the next pattern.
• A COMMIT will be issued for the bottom half of the device.
• The pass continues to erase the bottom half of the device.
• At the end of the last pass, the entire volume will be committed.