An efficient I/O interface for optical disks

Reviewer: Clement R. Attanasio

Optical disks are logically different form conventional magnetic disks in that they can be written on exactly one time. They can be thought of as initially having all bits set to zero; each bit can be set to one, after which it is forever equal to one. Within this constraint, how can (or should) such a device be used for a traditional data storage application__?__ The attractive attribute of optical disks is their increased storage capacity (by “orders of magnitude”) per unit cost, with comparable performance. This paper claims a slightly different point of view from previous work in the use of optical disks. The author asserts that it is more reasonable to investigate the use of optical disks for environments which take advantage of their characteristics, rather than viewing them as one-for-one substitutes for magnetic disks. Hence, he asserts that optical disks should be considered for environments in which update activity is not heavy. Furthermore, they have a strong natural advantage for an application which requires that previous values of data must be kept even after update. The main body of the paper presents and analyzes algorithms which the author claims are appropriate, i.e., “disk-efficient” and “optimal” in performance, under these assumptions. The paper is quite long and perhaps longer than necessary. Some of the analysis of suboptimal algorithms could have been omitted. The author's conclusion that he has presented “a practical approach for allowing operating systems and application programs that currently use magnetic disks to use optical disks instead” seems overstated, given his original approach that optical disks are not appropriate for high-update environments (what operating system does not write temporary files__?__). Although several of the author's statements can be disputed, I feel the time needed to read this paper would be well spent.