The Design of a High-Integrity Disk Management Subsystem
This dissertation describes and experimentally evaluates the design of the Logical Disk, a disk management subsystem that guarantees the integrity of data stored on disk even after system failures, while still providing performance competitive to other storage systems. Current storage systems that use the hard disk as storage medium, such as file systems, often do not provide sufficient protection against loss of data after a system failure. The designers of such systems are afraid that the amount of effort necessary for data protection would also result in too much loss of performance. The Logical Disk uses many different techniques to guarantee data integrity, including the support to execute multiple commands as one atomic action and avoiding `in-place updates' at all times. The techniques used to provide competitive performance include the technique of combining many, small write commands into one large, sequential, and thus efficient, write to disk, and clustering the data on disk continuously and automatically.