6.830: Lecture 17 (11/12/09)

We will continue our discussion of parallel and distributed query processing. You should read:

This paper discusses distributed transactions, addressing the problem of providing ACID-style semantics in a shared nothing environment.

As you read the papers, consider the following questions:

In the "R*" paper, how does the two phase commit (2PC) protocol work? What problem does it solve? What are the costs of using it?
What is the significance of the Presumed Abort/Presumed Commit variants of 2PC? How do they reduce the overhead of 2PC? When should you choose one over the other?