Computation Slicing: Techniques and Theory

We generalize the notion of slice introduced in our earlier paper [6]. A slice of a distributed computation with respect to a global predicate is the smallest computation that contains all consistent cuts of the original computation that satisfy the predicate. We prove that slice exists for all global predicates. We also establish that it is, in general, NP-complete to compute the slice. An optimal algorithm to compute slices for special cases of predicates is provided. Further, we present an efficient algorithm to graft two slices, that is, given two slices, either compute the smallest slice that contains all consistent cuts that are common to both slices or compute the smallest slice that contains all consistent cuts that belong to at least one of the slices. We give application of slicing in general and grafting in particular to global property evaluation of distributed programs. Finally, we show that the results pertaining to consistent global checkpoints [14, 18] can be derived as special cases of computation slicing.