Comment by tmulcahy

Any details/links that explain this? The article seems to suggest that they used to detect the deadlock as it was about to happen, and then abort everything and retry. This doesn't seem too different from "when they happen". What is the optimization?