Review

By the late 1970's and early 80's, there were a handful of successful transactional database systems that had been built, IMS, System R, Ingres, Oracle. However, the magic by which these systems implemented transactions was, with the exception of one paper, hidden deep in the depths of the system implementations. That exception was the "Notes" paper written by Jim Gray. Here, for the first time, were brought together in one place commit protocols, concurrency control, and recovery. Indeed, before this paper, there was excruciatingly little on recovery in the literature. In this one paper it was possible to learn about write-ahead logging, the DO-REDO-UNDO log record protocol, recovery checkpointing, restart, several flavors of two phase commit, multi-granularity locking, strict two phase locking, etc. In addition, one could learn about recoverable messages, record management, and a variety of systems issues. The paper not only described solutions to hard problems, it provided insights into what was important and why.

I avidly read the "Notes" paper when it came out (and I was surely not alone) and returned to it again and again over the years. The paper is a bit dated now, superceded partially by the System R recovery paper, the Bernstein, Hadzilacos, and Goodman text book, and fully by the monumental book authored by Jim himself (together with Andreas Reuter). But the "Notes" paper, which did not appear in either a conference or a journal, was the single most useful paper in the database literature for an entire generation of transaction system researchers and implementors, myself among them. Indeed, my interest in recovery dates from the reading of the "Notes" paper, so it continues to have an influence on my research directions to this day.