In a continuous query environment, different applications may have distinct Quality of Service (QoS) requirements. Given the unpredictability of streaming data, utilizing a single scheduling algorithm, as done by current state-of-the-art stream query engines, is no longer sufficient. Current scheduling algorithms used in these systems are typically one-dimensional, limiting the ability to perform well under changing system conditions. We propose a novel algorithm selection framework used in our CAPE system. This framework leverages the strengths of current scheduling algorithms to meet sets of QoS requirements. In CAPE, each algorithm can be compared in terms of its past ability to improve the QoS, knowing nothing about the characteristics of the algorithm. This knowledge can be used to determine the algorithm that probabilistically has the best chance of improving the QoS. Our framework has the flexibility to add new algorithms, query plans and data sets during runtime, with no need to fine-tune the algorithm to the system. Using standard data sets and query plans from existing literature on scheduling, our experiments show in fact that, this new framework combines the relative strengths of these algorithms while adhering to given QoS requirements.
Sutherland, Timothy M.
, Pielech, Bradford
, Rundensteiner, Elke A.
(2004). Adaptive QoS-Driven Scheduling Framework for a Continuous Query System. .
Retrieved from: http://digitalcommons.wpi.edu/computerscience-pubs/69