Exploring Failure Transparency and the Limits of Generic Recovery
David E. Lowell, Subhachandra Chandra, Peter M. Chen

This wonderful paper describes the goal and mechanisms of fault transparency 
or how a system may fail and recover without outside intervention or a
noticeable loss of performance. Based on modest experimentation, the authors
conclude that most types of faults preclude fault transparency.

The principal contribution by the authors is an excellent discussion of the
save-work and lose-work theorems which are required for a consistent failure
recovery. The paper lays the groundwork for reasoning about both fixed and
transient non-deterministic events within the context of how systems must
commit in order to maintain save-work without propagating failures. The
discussion concludes with an analysis of the costs of upholding save-work and
conflicts between the two invariants.

The papers chief weakness lies in the authors experimentations which are not
robust enough to support their findings but which are obvious in light of the
authors excellent reasoning. However, the incomplete attempt to measure the
invariant conflicts should not detract from the paper but perhaps offer an
incentive to others to perform better experiments.

This paper is both timely and timeless, and should be required reading for any
study of fault-tolerance. The well explained conflict between the save- and
lose-work theorems is vital to understanding and perhaps minimizing crash
events. The authors dangerous path algorithms offer interesting avenues for
future research in automated software verification.