147Incremental MapReduce Computations
memoization techniques for iterative computations. Incoop improves on these pro-
posals using a set of principles from related work to identify and overcome the situa-
tions where task-level memoization is inefcient, such as the stable input partitioning
or the contraction phase.
Our own short position paper [10] makes the case for applying techniques
inspired by self-adjusting computation to large-scale data processing in general and
uses MapReduce as an example. This position paper, however, models MapReduce
in a sequential, single-machine implementation of self-adjusting computation called
CEAL [24] and does not offer a full-scale distributed design and implementation
such as the system we presented.