5. Scheduling for Operation-Transfer Updates

Under the operation-transfer update model, the propagated updates involve partial content rather than the entire value. Hence, each replica basically reconstructs the current value of a data object from a history of propagated updates. This manner is suited for key-value stores such as PNUTS, Cassandra, and BigTable with a schema-like structured values where the value component is divided into columns as in traditional row structures and the system is responsible for the semantic interpretation of the read and write operations. Under operation-transfer updates, each data object in key-value stores accessed by its key still leads to a clear relationship between the arriving queries and their corresponding pending updates. In this chapter (Part of this chapter are reprinted from Xu et al., Distrib Parallel Databases 32(4): 535–581, 2014 [1], with kind permission from Springer Science+Business Media.), based on an operation-transfer model for update propagation, we present scheduling strategies for the efficient processing of both pending queries and updates at key-value data store nodes. In the following, Sect. 5.1 describes hybrid on-demand (HOD) mechanism; Sect. 5.2 presents freshness/tardiness (FIT) mechanism; Sect. 5.3 introduces popularity-aware mechanism; Sect. 5.4 shows the experimental analysis based on a simulation platform; Sect. 5.5 summarizes this chapter.